From 32a0c0db7ac87987567b288780c2106634a798ee Mon Sep 17 00:00:00 2001
From: Cedric Karungu <ckarungu921@gmail.com>
Date: Tue, 6 Oct 2026 17:56:36 +0200
Subject: [PATCH 1/5] feat(local): encode a podcast episode as MP3

audio.cpp's joined WAV is encoded in-process by LAME through lameenc, at
64 kbit/s per channel, the model's channels and rate. Without lameenc the
episode stays WAV. The frozen worker ships lameenc's dist-info for LAME's
LGPL text.

Fixes #2001
---
 docs/architecture/packaging.md                |  1 +
 docs/architecture/studio.md                   |  7 +--
 surfsense_local/backend/bundling/worker.spec  |  5 ++
 .../llm/providers/audiocpp/encoded_episode.py | 44 +++++++++++++++
 .../llm/providers/audiocpp/joined_wav.py      |  4 +-
 .../modules/llm/providers/audiocpp/speech.py  |  3 +-
 surfsense_local/backend/pyproject.toml        |  1 +
 .../llm/providers/audiocpp/test_speech.py     | 41 ++++++++++++--
 .../tests/unit/worker/test_studio_podcast.py  | 29 ++++++++--
 surfsense_local/backend/uv.lock               | 54 +++++++++++++++++++
 10 files changed, 175 insertions(+), 14 deletions(-)
 create mode 100644 surfsense_local/backend/modules/llm/providers/audiocpp/encoded_episode.py

diff --git a/docs/architecture/packaging.md b/docs/architecture/packaging.md
index 7d4b400b1..1bf6d1685 100644
--- a/docs/architecture/packaging.md
+++ b/docs/architecture/packaging.md
@@ -47,6 +47,7 @@ What else each spec names, and why the analyser cannot find it on its own:
 | `worker.spec` | python-docx, python-pptx, xlsxwriter, reportlab, openpyxl | the Office formats and document scripts run model-written code that imports them, so no static import exists; openpyxl, named in `pyproject.toml` though Docling brings it too, also reads every document script's workbook for its summary |
 | `worker.spec` | `matplotlib.pyplot`, `matplotlib.backends.backend_pdf`, `matplotlib.backends.backend_svg`, with the matplotlib hook set to the `Agg` backend | document scripts draw charts, and nothing imports matplotlib statically. Naming pyplot runs PyInstaller's hook, which adds `mpl-data` and the backend chosen; a chart saved as PDF or SVG, or `PdfPages`, imports its canvas by name, which the hook does not follow. `Agg` alone keeps Tk's GUI backend out, since scripts run with `MPLBACKEND=Agg` |
 | `worker.spec` | `modules.documents.tasks`, `modules.artifacts.tasks` | Huey resolves a task by its name |
+| `worker.spec` | `copy_metadata("lameenc")` | a podcast is encoded as MP3 by LAME, statically linked into `lameenc`'s extension, which the analyser finds by its import ([studio](studio.md#voicing-a-podcast)); LAME's LGPL licence text is only in the wheel's dist-info, which ships only when named |
 
 ## Model packs
 
diff --git a/docs/architecture/studio.md b/docs/architecture/studio.md
index 8ae35932b..a14feab5a 100644
--- a/docs/architecture/studio.md
+++ b/docs/architecture/studio.md
@@ -34,7 +34,7 @@ The catalog is a tuple of twelve `Format` rows in [`formats.py`](../../surfsense
 | `mindmap` | Mind map | text_gen | `content/mindmap/` | none; the outline is the body |
 | `flashcards` | Flashcards | text_gen | `content/flashcards/` | deck JSON |
 | `quiz` | Quiz | text_gen | `content/quiz/` | quiz JSON |
-| `podcast` | Podcast | text_gen, audio_gen | `media/audio/podcast/` | WAV audio |
+| `podcast` | Podcast | text_gen, audio_gen | `media/audio/podcast/` | MP3 audio, or WAV from a server that sends WAV |
 | `image` | Image | image_gen, text_gen | `media/visual/image/` | the image |
 | `infographic` | Infographic | image_gen, text_gen | `media/visual/infographic/` | the image |
 
@@ -82,7 +82,7 @@ Studio's Word and PDF buttons take one of two paths, chosen by [`office/document
 - The job sends the whole spec, fenced, and the instruction in one call, gets the whole revised spec back, and renders it as a draft of that kind would. Nothing is patched or diffed. The rewritten spec is kept only when it renders: a failed refine keeps none, is not retried by Huey, and leaves every earlier version as it was.
 - The agent's `surfsense_read_document` and `surfsense_render_document` refuse any version of a document Studio made, with "This document was made in Studio: refine it there, or make a new document." ([07](../proposals/file-agent/07-create-and-edit-mvp.md), decision 8).
 
-Studio uses no Electron `printToPDF`, which only draws the agent's Word and PowerPoint previews and its sources' pages ([agent](agent.md#previews)), and there is no ffmpeg in `surfsense_local`. A PDF is ReportLab through the builder, a Studio script or a script document, and a podcast is one `audio/wav` joined from each turn's WAV, or one `audio/mpeg` from a server that sends only MP3.
+Studio uses no Electron `printToPDF`, which only draws the agent's Word and PowerPoint previews and its sources' pages ([agent](agent.md#previews)), and there is no ffmpeg in `surfsense_local`. A PDF is ReportLab through the builder, a Studio script or a script document. A local podcast is one `audio/mpeg` encoded by LAME, through the `lameenc` package, from each turn's WAV joined; a server's is one `audio/wav` joined from its WAV turns, or one `audio/mpeg` from a server that sends only MP3.
 
 ## Script documents
 
@@ -124,6 +124,7 @@ The libraries a script may use are the worker's: python-docx, ReportLab, python-
 - **The voices are the model's roster**, from its manifest entry: each voice with the languages it speaks, one for a Kokoro or Kitten voice and all 31 for a Supertonic voice. The brief's language list, A to Z by the name shown in the interface language, and each speaker's voice picker come from it, the picker grouped into female and male voices. The brief opens in American English where a voice speaks it, else in any English, else in the roster's first language, and a remembered brief the chosen model cannot voice falls back to that.
 - **Memory is checked before anything loads, twice.** The worker compares the operating system's available memory (`MemAvailable` on Linux, free plus inactive pages on macOS, `GlobalMemoryStatusEx` on Windows) with the model's peak measured while voicing with every chunk of text full, from its entry, plus 1 GiB, and refuses short of it: "Voicing needs about 3.5 GB free; this computer has 1.8 GB. Supertonic 3 needs about 1.6 GB." The second sentence names the first other curated audio model, in the manifest's order, that would fit, and is left out when none would. It checks first, before the chat model drafts anything, so a machine that cannot voice the episode does not spend minutes writing it. Short there, it gives up the text runtime through the API ([the image job's hold](local-models/admission.md#the-image-jobs-hold)), which lets running replies finish and unloads the chat model, checks again, and lets it go for drafting, because voicing frees that memory anyway: on 16 GB, Qwen3 1.7B resident left 3.3 GB free and refused Kokoro's 3.5 before a word was drafted. At voicing it gives the runtime up again and holds it until the last turn is voiced, so a chat sent meanwhile waits instead of loading the chat model back into memory voicing needs. The router reloads the chat model on the next request after either. Each check waits up to 10 seconds for memory the app is giving back before it refuses, because a starting podcast ends sd-server's idle minutes but Electron stops it only on its next 5-second poll ([the image path](#the-image-path)). audio.cpp's own guard counts only the file it reads.
 - **One speech request per turn**, with the turn's voice. The brief's language goes in the request only for a voice that speaks several; a Kokoro voice's name already fixes its language. The turns are joined with 0.35 s between them.
+- **The episode is MP3**: the joined 16-bit WAV is encoded in-process by LAME through `lameenc`, at 64 kbit/s per channel and the model's own channels and rate, about a sixth of the WAV ([`encoded_episode.py`](../../surfsense_local/backend/modules/llm/providers/audiocpp/encoded_episode.py)). No ffmpeg and nothing on the user's `PATH` is used. Where `lameenc` cannot be imported, as in a checkout that skipped it, the episode stays WAV, which the viewer's `<audio>` plays the same.
 - **A turn the server fails ends the episode with the server's own words**: "audio.cpp could not voice turn 3 of 40: unknown Kokoro voice id: nobody". The server was reached, so this is not "The model could not be reached", which stays for a server that does not answer.
 - **The model is unloaded when voicing ends**, with `POST /v1/tasks/unload_all_models`, success or failure, so its memory is back before the chat model's next job. The server's five-minute idle unload is the backstop.
 
@@ -259,4 +260,4 @@ Every pipeline returns a `Built`: a `title`, the `markdown` that is always the i
 - A document script runs with the user's privileges: the runner gives it a clean environment and a time limit, but it can read any file the user can and open network connections ([ADR 0039](../adr/0039-document-scripts-run-without-approval.md)).
 - A failed script's stored reason repeats the run folder's absolute path in every traceback frame, about 130 characters each against its 500, so a long traceback keeps few of its lines for the agent.
 - A workbook's formulas show the value the script stored with them, since neither the runner nor Studio's viewer recalculates: xlsxwriter stores 0 unless given one, which the skill tells the agent to pass. Excel recalculates when it opens the file.
-- A podcast is WAV. The design encodes MP3 with a bundled ffmpeg, which is not built.
+- A podcast voiced on a server that sends WAV stays WAV: only audio.cpp's episode goes through the MP3 encoder.
diff --git a/surfsense_local/backend/bundling/worker.spec b/surfsense_local/backend/bundling/worker.spec
index 58cbfb89b..5af9ca268 100644
--- a/surfsense_local/backend/bundling/worker.spec
+++ b/surfsense_local/backend/bundling/worker.spec
@@ -13,6 +13,7 @@ from PyInstaller.utils.hooks import (
     collect_all,
     collect_data_files,
     collect_dynamic_libs,
+    copy_metadata,
 )
 
 sys.path.insert(0, SPECPATH)
@@ -84,6 +85,10 @@ binaries += collect_dynamic_libs("tokenizers")
 datas += collect_data_files("worker.studio", includes=["**/*.md"])
 datas += collect_data_files("modules.chat", includes=["prompts/*.md"])
 
+# lameenc, which encodes a podcast as MP3, is found by its import; the wheel
+# carries LAME's LGPL text only in its dist-info, which ships only when named.
+datas += copy_metadata("lameenc")
+
 # Huey resolves a task by its name, so the module that registers it must be in.
 hiddenimports += ["modules.documents.tasks", "modules.artifacts.tasks"]
 
diff --git a/surfsense_local/backend/modules/llm/providers/audiocpp/encoded_episode.py b/surfsense_local/backend/modules/llm/providers/audiocpp/encoded_episode.py
new file mode 100644
index 000000000..fbeccc055
--- /dev/null
+++ b/surfsense_local/backend/modules/llm/providers/audiocpp/encoded_episode.py
@@ -0,0 +1,44 @@
+"""The joined episode as MP3, about a sixth of its WAV, through LAME (lameenc).
+WAV where lameenc is not installed, so a checkout without it still plays."""
+
+import io
+import logging
+import wave
+
+from modules.llm.providers.protocols import SynthesizedAudio
+
+__all__ = ["encoded_episode"]
+
+logger = logging.getLogger(__name__)
+
+# Per channel: speech at 24 kHz mono stays clear at 64 kbit/s.
+KBPS_PER_CHANNEL = 64
+# LAME's 0 (best) to 9 (fastest); 2 is its recommended near-best.
+QUALITY = 2
+
+
+def encoded_episode(wav: bytes) -> SynthesizedAudio:
+    """`wav` is joined_wav's output: one format, 16-bit PCM from audio.cpp."""
+    try:
+        import lameenc
+    except ImportError:
+        logger.warning("audiocpp: lameenc is not installed; the episode stays WAV")
+        return SynthesizedAudio(wav, "audio/wav")
+    with wave.open(io.BytesIO(wav)) as pcm:
+        channels, width, rate = (
+            pcm.getnchannels(),
+            pcm.getsampwidth(),
+            pcm.getframerate(),
+        )
+        frames = pcm.readframes(pcm.getnframes())
+    # LAME takes 16-bit samples only.
+    if width != 2:
+        return SynthesizedAudio(wav, "audio/wav")
+    encoder = lameenc.Encoder()
+    encoder.set_channels(channels)
+    encoder.set_in_sample_rate(rate)
+    encoder.set_bit_rate(KBPS_PER_CHANNEL * channels)
+    encoder.set_quality(QUALITY)
+    return SynthesizedAudio(
+        bytes(encoder.encode(frames) + encoder.flush()), "audio/mpeg"
+    )
diff --git a/surfsense_local/backend/modules/llm/providers/audiocpp/joined_wav.py b/surfsense_local/backend/modules/llm/providers/audiocpp/joined_wav.py
index 7054d5d7e..81d584285 100644
--- a/surfsense_local/backend/modules/llm/providers/audiocpp/joined_wav.py
+++ b/surfsense_local/backend/modules/llm/providers/audiocpp/joined_wav.py
@@ -1,5 +1,5 @@
-"""One WAV from each turn's, with a pause between speakers. WAV out: the
-browser plays it and the standard library writes it, with no encoder to ship.
+"""One WAV from each turn's, with a pause between speakers. WAV, because a
+pause is silent frames, which only PCM takes; encoded_episode makes it MP3.
 """
 
 import io
diff --git a/surfsense_local/backend/modules/llm/providers/audiocpp/speech.py b/surfsense_local/backend/modules/llm/providers/audiocpp/speech.py
index b38eaf23d..05743573a 100644
--- a/surfsense_local/backend/modules/llm/providers/audiocpp/speech.py
+++ b/surfsense_local/backend/modules/llm/providers/audiocpp/speech.py
@@ -11,6 +11,7 @@ import httpx
 
 from modules.llm.catalog.local.engines.audiocpp.manifest_fields import AudioDefaults
 from modules.llm.hardware import system_memory
+from modules.llm.providers.audiocpp.encoded_episode import encoded_episode
 from modules.llm.providers.audiocpp.joined_wav import joined_wav
 from modules.llm.providers.audiocpp.memory import (
     NotEnoughMemoryError,
@@ -108,7 +109,7 @@ class AudioCppSpeech:
                     voiced = await self._voice_each(client, turns, language, speaks)
                 finally:
                     await _unload(client)
-        return SynthesizedAudio(joined_wav(voiced), "audio/wav")
+        return encoded_episode(joined_wav(voiced))
 
     async def _wait_for_memory(self, seconds: float) -> None:
         deadline = time.monotonic() + seconds
diff --git a/surfsense_local/backend/pyproject.toml b/surfsense_local/backend/pyproject.toml
index de0276930..884619bfa 100644
--- a/surfsense_local/backend/pyproject.toml
+++ b/surfsense_local/backend/pyproject.toml
@@ -53,6 +53,7 @@ dependencies = [
     "torchvision>=0.17",
     "uvicorn[standard]>=0.34.0",
     "xlsxwriter>=3.2.9",
+    "lameenc>=1.8.4",
 ]
 
 [dependency-groups]
diff --git a/surfsense_local/backend/tests/unit/llm/providers/audiocpp/test_speech.py b/surfsense_local/backend/tests/unit/llm/providers/audiocpp/test_speech.py
index 6d9f206ee..ad39995f0 100644
--- a/surfsense_local/backend/tests/unit/llm/providers/audiocpp/test_speech.py
+++ b/surfsense_local/backend/tests/unit/llm/providers/audiocpp/test_speech.py
@@ -3,6 +3,7 @@
 import asyncio
 import io
 import json
+import sys
 import wave
 from collections.abc import AsyncIterator
 from contextlib import asynccontextmanager
@@ -62,10 +63,14 @@ def test_the_voices_are_the_chosen_models_roster() -> None:
     """A Supertonic voice speaks every language the model does; a Kokoro voice
     speaks its own."""
     kokoro = AudioCppSpeech(
-        voiced("kokoro-82m", "kokoro-82m-q8_0"), base_url="", give_up_text_runtime=NO_CHAT
+        voiced("kokoro-82m", "kokoro-82m-q8_0"),
+        base_url="",
+        give_up_text_runtime=NO_CHAT,
     )
     supertonic = AudioCppSpeech(
-        voiced("supertonic-3", "supertonic-3-f16"), base_url="", give_up_text_runtime=NO_CHAT
+        voiced("supertonic-3", "supertonic-3-f16"),
+        base_url="",
+        give_up_text_runtime=NO_CHAT,
     )
 
     heart = next(v for v in kokoro.voices() if v.id == "af_heart")
@@ -145,8 +150,10 @@ def speak(model: VoicedModel, server: StubServer, turns, language="en-US", free=
     return asyncio.run(speech.synthesize(turns, language))
 
 
-def test_each_turn_is_one_request_and_the_turns_join_with_a_pause() -> None:
-    """0.35 s between speakers, as the podcast has always had, at the model's rate."""
+def test_each_turn_is_one_request_and_the_turns_join_with_a_pause(monkeypatch) -> None:
+    """0.35 s between speakers, as the podcast has always had, at the model's rate.
+    Read from the WAV, which an episode stays when no encoder is installed."""
+    monkeypatch.setitem(sys.modules, "lameenc", None)
     server = StubServer(frames=1000)
     turns = [SpokenTurn("am_adam", "Welcome back."), SpokenTurn("af_heart", "Thanks.")]
 
@@ -163,6 +170,32 @@ def test_each_turn_is_one_request_and_the_turns_join_with_a_pause() -> None:
         assert joined.getnframes() == 10_400
 
 
+def test_the_episode_is_mp3_at_the_models_rate() -> None:
+    """About a sixth of the WAV, and the browser plays it as it did the WAV."""
+    turns = [SpokenTurn("am_adam", "Welcome back."), SpokenTurn("af_heart", "Thanks.")]
+
+    audio = speak(voiced("kokoro-82m", "kokoro-82m-q8_0"), StubServer(24000), turns)
+
+    assert audio.media_type == "audio/mpeg"
+    header = audio.content[:4]
+    assert header[0] == 0xFF and header[1] & 0xE0 == 0xE0, "an MPEG frame first"
+    # MPEG-2's rate index 1 is 24 kHz, mono is channel mode 3.
+    assert (header[1] >> 3 & 0b11, header[2] >> 2 & 0b11) == (0b10, 1)
+    assert header[3] >> 6 == 0b11
+    assert len(audio.content) < len(wav(2 * 24000)) / 4
+
+
+def test_the_episode_stays_wav_without_an_encoder(monkeypatch) -> None:
+    """A checkout that skipped the dependency still gets an episode it can play."""
+    monkeypatch.setitem(sys.modules, "lameenc", None)
+    turns = [SpokenTurn("am_adam", "Welcome back."), SpokenTurn("af_heart", "Thanks.")]
+
+    audio = speak(voiced("kokoro-82m", "kokoro-82m-q8_0"), StubServer(), turns)
+
+    assert audio.media_type == "audio/wav"
+    assert audio.content[:4] == b"RIFF"
+
+
 def test_the_language_is_sent_only_where_the_voice_speaks_several() -> None:
     """Supertonic reads it from the request; a Kokoro voice's name already
     fixes it, and Kokoro's codes differ from the manifest's."""
diff --git a/surfsense_local/backend/tests/unit/worker/test_studio_podcast.py b/surfsense_local/backend/tests/unit/worker/test_studio_podcast.py
index ccbe73ebf..77f62c65f 100644
--- a/surfsense_local/backend/tests/unit/worker/test_studio_podcast.py
+++ b/surfsense_local/backend/tests/unit/worker/test_studio_podcast.py
@@ -229,9 +229,11 @@ def test_every_segment_reads_the_same_sources_before_what_is_its_own() -> None:
 class FakeVoice:
     """A TextToSpeech that records what it was asked to say."""
 
-    def __init__(self) -> None:
+    def __init__(self, content: bytes = b"RIFFfake", media_type: str = "audio/wav"):
         self.turns: list[SpokenTurn] = []
         self.language: str | None = None
+        self.content = content
+        self.media_type = media_type
 
     def voices(self) -> list[Voice]:
         return [
@@ -247,11 +249,13 @@ class FakeVoice:
     ) -> SynthesizedAudio:
         self.turns = turns
         self.language = language
-        return SynthesizedAudio(b"RIFFfake", "audio/wav")
+        return SynthesizedAudio(self.content, self.media_type)
 
 
-def _episode(monkeypatch: pytest.MonkeyPatch, *replies: str) -> tuple[FakeVoice, list]:
-    voice = FakeVoice()
+def _episode(
+    monkeypatch: pytest.MonkeyPatch, *replies: str, voice: FakeVoice | None = None
+) -> tuple[FakeVoice, list]:
+    voice = voice or FakeVoice()
     queue = iter(replies)
     prompts: list[str] = []
 
@@ -299,6 +303,23 @@ def test_an_episode_is_planned_then_drafted_per_segment_then_voiced_per_speaker(
     assert "**Lee:** Goodbye." in built.markdown
 
 
+def test_an_mp3_episode_is_named_mp3(monkeypatch: pytest.MonkeyPatch) -> None:
+    """The local voice encodes MP3; the stored file says so by its name."""
+    voice, _ = _episode(
+        monkeypatch,
+        '{"title": "Saturn Rings", "segments": [{"title": "Open"}]}',
+        '{"turns": [{"speaker": 1, "text": "Hi."}, {"speaker": 2, "text": "Bye."}]}',
+        voice=FakeVoice(b"\xff\xf3fake", "audio/mpeg"),
+    )
+
+    built = pipeline.render(MODEL, voice, [], None, BRIEF.model_dump(mode="json"))
+
+    assert (built.primary_mime, built.primary_filename) == (
+        "audio/mpeg",
+        "saturn-rings.mp3",
+    )
+
+
 @pytest.mark.parametrize(
     ("speakers", "cast"),
     [
diff --git a/surfsense_local/backend/uv.lock b/surfsense_local/backend/uv.lock
index 1f0682f4e..9031a5c12 100644
--- a/surfsense_local/backend/uv.lock
+++ b/surfsense_local/backend/uv.lock
@@ -1233,6 +1233,58 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/a6/c1/69f00d627949580e43d57af0aa465df46868d7c29801c137a55374101294/kiwisolver-1.5.1-graalpy312-graalpy250_312_native-win_amd64.whl", hash = "sha256:a5a00665d1a0e26763a7338d7e911d4598fbc1d50dd0d6b7919b7dc6c5d6569f", size = 73362, upload-time = "2026-08-28T10:28:12.449Z" },
 ]
 
+[[package]]
+name = "lameenc"
+version = "1.8.4"
+source = { registry = "https://pypi.org/simple" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/e7/41/afa8b9bd15ebe757b8a1029b1f44b0caa94252dd12537d78a81c360ad069/lameenc-1.8.4-cp312-cp312-macosx_10_13_universal2.whl", hash = "sha256:8482f68a0910606efc182f1858fef8655681d9d29c8edc9fa5c36acf74819118", size = 189628, upload-time = "2026-06-27T15:03:08.34Z" },
+    { url = "https://files.pythonhosted.org/packages/4b/bd/d64e49025090c1971eb085076a40d82f3fcd8339f2a2a4e1e224bd9aa482/lameenc-1.8.4-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:fb0d5bb76b09d8bf4e27f4824a72e4acd659bd4ec8dac2879fd5744f3d6d88fc", size = 178226, upload-time = "2026-06-27T15:02:59.939Z" },
+    { url = "https://files.pythonhosted.org/packages/a3/1a/fa4d2e4df30b6322a806da58b214c5c8de30e4136027dddbbaa1238e5c1c/lameenc-1.8.4-cp312-cp312-manylinux1_i686.manylinux_2_34_i686.manylinux_2_5_i686.whl", hash = "sha256:43500c41c51a88bdca9b4ee85c5764d4c0d8c5b1d1cb9cc35c2449fc2e0412f9", size = 221785, upload-time = "2026-06-27T15:02:46.71Z" },
+    { url = "https://files.pythonhosted.org/packages/7f/80/9f1ae88f9dc02b6a9ad53ab687c3e13077bb81f3f452bb59f17b42318ba0/lameenc-1.8.4-cp312-cp312-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:ea7a7968b20535934bc11caca3d23b12e972de6e02f31bdc6a9e206c198cfd1e", size = 251884, upload-time = "2026-06-27T15:12:18.347Z" },
+    { url = "https://files.pythonhosted.org/packages/98/4a/f5856aa2362feb8afc1a9e51d81a946b82413b465f5577943984dafab256/lameenc-1.8.4-cp312-cp312-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_34_aarch64.whl", hash = "sha256:606ee90e18b70b0134c410fe21db11e31bc539e1da1a2c298d90889878766552", size = 251631, upload-time = "2026-06-27T15:07:58.681Z" },
+    { url = "https://files.pythonhosted.org/packages/49/98/ced7da98fb0c149e80d3a5a97546b5abcec6a06f4187cc8842a737107487/lameenc-1.8.4-cp312-cp312-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:00d619c0a617f66feccbbd2fa9ed3857958ea503f9fe0038cb8b1d950b8b6452", size = 247546, upload-time = "2026-06-27T14:58:55.928Z" },
+    { url = "https://files.pythonhosted.org/packages/48/03/1d153252a5aa9093a461b3d013b1e8d383806f6c8c59c7f65c6928197aaa/lameenc-1.8.4-cp312-cp312-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_34_x86_64.whl", hash = "sha256:18ba38c49759e217dd6fecf56ef92eab2a24f0a0d87ae4c3564ce4748d75b166", size = 247480, upload-time = "2026-06-27T15:02:59.079Z" },
+    { url = "https://files.pythonhosted.org/packages/24/5c/f7f73b6ed2a46d149b7f8a2046c26e61e2cd4ac248f628cebce300abbf31/lameenc-1.8.4-cp312-cp312-win32.whl", hash = "sha256:513b5163b30581350be6c3e6adb58fd63ab1573ee534f5e9270655f3ffe63562", size = 124729, upload-time = "2026-06-27T15:03:41.39Z" },
+    { url = "https://files.pythonhosted.org/packages/6e/d1/b4b08b1c27b4991052db2fae3082100a6317fef34873bbeb121809315b22/lameenc-1.8.4-cp312-cp312-win_amd64.whl", hash = "sha256:33854f5b479cec81679860c8d67225e2ab3a31a0bde0bdf49b55e2bd6ee1923e", size = 153045, upload-time = "2026-06-27T15:03:40.24Z" },
+    { url = "https://files.pythonhosted.org/packages/8b/bd/ccbf35970373ab076e5036c1f14670eb13ed05bf4dbb2fbdfefe35e8b812/lameenc-1.8.4-cp312-cp312-win_arm64.whl", hash = "sha256:e72e10ea0240bcc46e05df9dd4979116e74e183a5983cd0dcb14ff5315444649", size = 131452, upload-time = "2026-06-27T15:03:36.726Z" },
+    { url = "https://files.pythonhosted.org/packages/f9/9c/608f3e1daf71203037fb3c04ff714fd369363d44d3a54d7fe21dbd307025/lameenc-1.8.4-cp313-cp313-macosx_10_13_universal2.whl", hash = "sha256:78d8cdb3175e7c55a34c705c101a9e6483ae18572be22a6066aa4ef359df68f7", size = 189620, upload-time = "2026-06-27T15:03:07.299Z" },
+    { url = "https://files.pythonhosted.org/packages/bf/f7/be59571f5ad29ad9a02d2e3fb69668ae06f5ea9ae1742fcda656e58de62f/lameenc-1.8.4-cp313-cp313-macosx_11_0_arm64.whl", hash = "sha256:05f1034b40d139a043c0ec877e968230dbc0945f320427d662d457277ab9bc4a", size = 183358, upload-time = "2026-06-27T15:03:04.675Z" },
+    { url = "https://files.pythonhosted.org/packages/fd/62/70c196a516b38bf7fb3529e1c7621dbe8b05cf12c53eecda2628d9e5234d/lameenc-1.8.4-cp313-cp313-manylinux1_i686.manylinux_2_34_i686.manylinux_2_5_i686.whl", hash = "sha256:f3279d497a21395378e30cbf632bd40606c292e0f39d152e237ffb429cab3c8b", size = 221854, upload-time = "2026-06-27T15:02:48.374Z" },
+    { url = "https://files.pythonhosted.org/packages/4e/1c/3a5863b8c8e2051ecce855a38099757218f8c0b2a2515015ce93519ff134/lameenc-1.8.4-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:d9ce4baad7f0516682a91aa11d1e8483fe1996640c9a8c0e667ec3aec65a6fc4", size = 251967, upload-time = "2026-06-27T15:12:19.877Z" },
+    { url = "https://files.pythonhosted.org/packages/e3/f6/ef38b5233ebddc05bae9ef5fe31e909f422b41a5a8e45073133fbf8fe191/lameenc-1.8.4-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_34_aarch64.whl", hash = "sha256:c3496f6e68fc6441b0f6972acab9298de85c2013f888f80dd69c41a4976470bc", size = 251714, upload-time = "2026-06-27T15:08:00.185Z" },
+    { url = "https://files.pythonhosted.org/packages/21/ab/61087872800c15f91c5e50c5331b139c4b55b62cbb3fbd25aeb87052e752/lameenc-1.8.4-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:0e5b46a8e4ebf3dd495afc05fc8efcda24eac17b386e2c60b0d2e708d266c154", size = 247632, upload-time = "2026-06-27T14:58:57.199Z" },
+    { url = "https://files.pythonhosted.org/packages/6c/2b/96dfcb4947b2fe558791009d621c831972b13dc3f4ab489d62585cd81d16/lameenc-1.8.4-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_34_x86_64.whl", hash = "sha256:7e08ab42b8b6c2467c386e1ebb62fec8dae00cbb25d803d25e14bffd46fc9087", size = 247592, upload-time = "2026-06-27T15:03:00.536Z" },
+    { url = "https://files.pythonhosted.org/packages/44/6c/d950bfcb0b8803ca281d5b72d1be7d0a045830ccda9a41fda86dd4c57669/lameenc-1.8.4-cp313-cp313-win32.whl", hash = "sha256:faf3926600c1f6ed577984e15647e5e459cedd9c929953acb70d605a2847b94e", size = 124725, upload-time = "2026-06-27T15:03:47.476Z" },
+    { url = "https://files.pythonhosted.org/packages/53/aa/673a0c57d2e7ae5d800a2a43024d5ac1660ee26c114149e26a4188be93c2/lameenc-1.8.4-cp313-cp313-win_amd64.whl", hash = "sha256:7db3df4133d7b39f2f09ad684bf0a7a92c2d11117a0afc5db5cb152e48025b63", size = 153036, upload-time = "2026-06-27T15:03:46.669Z" },
+    { url = "https://files.pythonhosted.org/packages/a8/23/5ade982d5d285b30144c7feb55a8680f2a883d14477046b44ec33c2cdac3/lameenc-1.8.4-cp313-cp313-win_arm64.whl", hash = "sha256:a9c40d7b054c2e8d816a95912268de52b7d3f5f1da250c73b611849c5159d072", size = 131448, upload-time = "2026-06-27T15:03:48.732Z" },
+    { url = "https://files.pythonhosted.org/packages/13/57/44735025842e06e5e00f37049585df1ab45922b91d874bcce85110dc9deb/lameenc-1.8.4-cp314-cp314-macosx_10_15_universal2.whl", hash = "sha256:55e468c75354fd3a1874282d4b23b605137025dca9b024bb8be8f4e91c5169e5", size = 189543, upload-time = "2026-06-27T15:03:23.256Z" },
+    { url = "https://files.pythonhosted.org/packages/97/d6/14e15129caa7cb4d2cb5c6b2b030d0bbc87ab9c7224be2a84d88997b3e78/lameenc-1.8.4-cp314-cp314-macosx_11_0_arm64.whl", hash = "sha256:859fa9f05e0c7e825efb72431f8243bcc4318c71ff3b4d57c7cebaed6fcadb65", size = 183350, upload-time = "2026-06-27T15:03:11.532Z" },
+    { url = "https://files.pythonhosted.org/packages/da/35/818502b8e55b4cd9f7743500015e9ce07f01d474acdade0b0bd5e5ad3221/lameenc-1.8.4-cp314-cp314-manylinux1_i686.manylinux_2_34_i686.manylinux_2_5_i686.whl", hash = "sha256:92dae11d2fd422c3c310900893edd3e20d538741959c7cd426d91af2cf18fe27", size = 222018, upload-time = "2026-06-27T15:02:49.686Z" },
+    { url = "https://files.pythonhosted.org/packages/de/9c/6fd42cb5c8fde74793042a16c3278a39c814f00ce37797ea61b642caeaff/lameenc-1.8.4-cp314-cp314-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:29fa3dfb57b3d1ef021b2c9b9b940e2502d139bcf0c84153cd0f57c4506b856d", size = 252091, upload-time = "2026-06-27T15:12:21.482Z" },
+    { url = "https://files.pythonhosted.org/packages/34/65/66211814595cd9ce2bbf8c7cea345c947fdae90f87573ccf082a2bbc525b/lameenc-1.8.4-cp314-cp314-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_34_aarch64.whl", hash = "sha256:627588bc0a2520b33e87d7966bedb1138b724f18c0a5d24a2a3a12de17351fad", size = 251856, upload-time = "2026-06-27T15:08:01.728Z" },
+    { url = "https://files.pythonhosted.org/packages/36/d6/224f9055296dfd16e44da364220167c2612402906fcc53e9e882d6bc72cc/lameenc-1.8.4-cp314-cp314-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:c6527a8ae8ac078010a1fecc697145e7be1bb163cd5092b5c32d4332b6430886", size = 247729, upload-time = "2026-06-27T14:58:58.417Z" },
+    { url = "https://files.pythonhosted.org/packages/43/6c/2298da206cdeac946cafc07d9e04d48e457874c96e6f5c8676cce39c83a0/lameenc-1.8.4-cp314-cp314-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_34_x86_64.whl", hash = "sha256:d44282c566712e42aee1624b5e406a9f277ae5395b729338bd20d844a98eb770", size = 247696, upload-time = "2026-06-27T15:03:02.216Z" },
+    { url = "https://files.pythonhosted.org/packages/5c/fa/30f3b02d8da0f209341b98a01f9918a7e2c68dee07949279a008f7f7647d/lameenc-1.8.4-cp314-cp314-win32.whl", hash = "sha256:31ab1bf3b191995293c1e085b43e3d78046341a156328d222d9a4d5eb3e149e3", size = 128577, upload-time = "2026-06-27T15:03:54.3Z" },
+    { url = "https://files.pythonhosted.org/packages/2c/c7/3132e584e9df8196013bd8104e17ca3247e18d5ebfe9c9369878fc8a5924/lameenc-1.8.4-cp314-cp314-win_amd64.whl", hash = "sha256:74ddfa8ba265924f958c1135dacc62345fcee05a9449b26a902541cfa9b9857e", size = 157385, upload-time = "2026-06-27T15:03:44.671Z" },
+    { url = "https://files.pythonhosted.org/packages/bf/a9/d88809cb11105984bd8fb17c98cae626f8ddf7a27e1e146fb89028cb81bf/lameenc-1.8.4-cp314-cp314-win_arm64.whl", hash = "sha256:d44397967f9b10daa3b6941d20e7035ec8d7c5168f1a108f831c6cd0de5ccd3c", size = 136928, upload-time = "2026-06-27T15:03:35.9Z" },
+    { url = "https://files.pythonhosted.org/packages/39/15/376104b0580a45e4da36c413850c979b58d602a9c2e08271deb40f4d5c89/lameenc-1.8.4-cp314-cp314t-manylinux1_i686.manylinux_2_34_i686.manylinux_2_5_i686.whl", hash = "sha256:239741e14b715676326a4b340fe475b6f1007ecc31d50c38fba96736536e26b0", size = 223798, upload-time = "2026-06-27T15:02:51.011Z" },
+    { url = "https://files.pythonhosted.org/packages/bf/3d/e693d99d943e0741780e917368152f8b0fe054311dbf6278ddb777bcd254/lameenc-1.8.4-cp314-cp314t-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:61ee4980f099b3791322591a150e2efe3468f5f2cf145af0c55c866f11708cf6", size = 254301, upload-time = "2026-06-27T15:12:23.095Z" },
+    { url = "https://files.pythonhosted.org/packages/64/1a/25fe55ae2a2e376c6ba1c40d4085ce2308ad32c125efc93d04b138c09639/lameenc-1.8.4-cp314-cp314t-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_34_aarch64.whl", hash = "sha256:694afa6da2d89856017493bb1089283293988ba6522bad28e23697850569335e", size = 253929, upload-time = "2026-06-27T15:08:03.077Z" },
+    { url = "https://files.pythonhosted.org/packages/32/af/668d9fc052852dd04fcb7093d55bd88484c2821bc0132adb75b017ede7c6/lameenc-1.8.4-cp314-cp314t-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:6a4948b98574c025e8902af0aaba905ce9bf032a0b6ce6a578b64817611860e5", size = 249448, upload-time = "2026-06-27T14:58:59.78Z" },
+    { url = "https://files.pythonhosted.org/packages/b7/d9/be995262968580b08e88e47d2e7a0192dce209009d554569a50a8335674c/lameenc-1.8.4-cp314-cp314t-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_34_x86_64.whl", hash = "sha256:970685ae4ac246dccc177e3dad16a27187582cd4cdc57208e894e6bf860699d7", size = 249276, upload-time = "2026-06-27T15:03:03.77Z" },
+    { url = "https://files.pythonhosted.org/packages/17/65/90a62ce8cb745daaab3883175cde26f19634b066c4305fbfabc0b1135ffa/lameenc-1.8.4-cp315-cp315-manylinux1_i686.manylinux_2_34_i686.manylinux_2_5_i686.whl", hash = "sha256:63bc671e3ca8a23654930af46251d848d67c4e47a566edd37f97795ce49bb82f", size = 222718, upload-time = "2026-06-27T15:02:52.234Z" },
+    { url = "https://files.pythonhosted.org/packages/4f/08/1f263ff43d4af81e62ad6c85401049f6519dd1e8539c349281c91e2235bd/lameenc-1.8.4-cp315-cp315-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:389b4210f47e68cf031c00db6f2cf41b517f6c5b00a8463e2c0dc1bbb3350394", size = 252497, upload-time = "2026-06-27T15:12:24.79Z" },
+    { url = "https://files.pythonhosted.org/packages/fa/7d/70f649abc1af4b9844c063dd51dcbec3e56b61373921d35e40976278c460/lameenc-1.8.4-cp315-cp315-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_34_aarch64.whl", hash = "sha256:e668d65f85b73d250b82c3a925c503c5b41ee3fe2ebff4255d620ebc8dff0148", size = 252265, upload-time = "2026-06-27T15:08:04.567Z" },
+    { url = "https://files.pythonhosted.org/packages/87/e1/2e4acbce8383b324d887eb24a78d2f9b815ee5a4c36fa6198b62d45c9664/lameenc-1.8.4-cp315-cp315-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:c00703b1c7fb7c2aecf453e750f0011914753ddbe529ec54ed98f53b8adba256", size = 248060, upload-time = "2026-06-27T14:59:00.926Z" },
+    { url = "https://files.pythonhosted.org/packages/c0/82/bb07020a50b140bdcc1a77c7f5b478560e80a50f58ca4b5feac85c059305/lameenc-1.8.4-cp315-cp315-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_34_x86_64.whl", hash = "sha256:1daa7739fb469558d2786909cee9e9e76a9f53fa93f8c1825acdc6c2d8192570", size = 248081, upload-time = "2026-06-27T15:03:05.229Z" },
+    { url = "https://files.pythonhosted.org/packages/25/c4/23f73c2a159083cccddbd4b69c1a59f2c97b239c4f8fe638daf753486a4b/lameenc-1.8.4-cp315-cp315t-manylinux1_i686.manylinux_2_34_i686.manylinux_2_5_i686.whl", hash = "sha256:08ec5c10472dd153a75b17a52b402b5d62628fa054d701fc4a27ddd86e037351", size = 224228, upload-time = "2026-06-27T15:02:53.54Z" },
+    { url = "https://files.pythonhosted.org/packages/e7/f5/00858b041b3dbdd833f3e28fed316bf73e2d0bc1519b560b5320af9679bd/lameenc-1.8.4-cp315-cp315t-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:0bf1463f79c7965922dd0604dfaedd636e9e74acefb21ec254419f2b42cf6a4e", size = 254707, upload-time = "2026-06-27T15:12:26.443Z" },
+    { url = "https://files.pythonhosted.org/packages/44/ae/3f091c3090e5dc093f781134a73d6ae41ad2f46426bfdb7bb1d884c47bae/lameenc-1.8.4-cp315-cp315t-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_34_aarch64.whl", hash = "sha256:2f8ae9b47b02c327ac4ab0f5378dafc1f7b5bf0bd30b90fa81033ee71f0005d8", size = 254307, upload-time = "2026-06-27T15:08:05.914Z" },
+    { url = "https://files.pythonhosted.org/packages/fb/f0/45a32cd5f41cc8462ecd97e47d651bf525e10ba1f7c71de3c5b18efcfdbc/lameenc-1.8.4-cp315-cp315t-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:b8aacb9f345ff0e6cab137d0d8436c5d5712b332c4998f42d167fb5677bee83e", size = 249855, upload-time = "2026-06-27T14:59:02.08Z" },
+    { url = "https://files.pythonhosted.org/packages/98/69/818d51a0c2004c26fd6c118eecb9508d6b672d22f813e11bf4586894be92/lameenc-1.8.4-cp315-cp315t-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_34_x86_64.whl", hash = "sha256:7f83753a35babf2e70d1d511c3fdde0ceedf1af4977b705cb591b8da47bb457d", size = 249696, upload-time = "2026-06-27T15:03:06.985Z" },
+]
+
 [[package]]
 name = "latex2mathml"
 version = "3.81.0"
@@ -2966,6 +3018,7 @@ dependencies = [
     { name = "gguf" },
     { name = "httpx" },
     { name = "huey" },
+    { name = "lameenc" },
     { name = "markdown-it-py" },
     { name = "matplotlib" },
     { name = "onnxruntime" },
@@ -3008,6 +3061,7 @@ requires-dist = [
     { name = "gguf", specifier = "==0.19.0" },
     { name = "httpx", specifier = ">=0.28.1" },
     { name = "huey", specifier = ">=3.3.4" },
+    { name = "lameenc", specifier = ">=1.8.4" },
     { name = "markdown-it-py", specifier = ">=4.0" },
     { name = "matplotlib", specifier = ">=3.11.2" },
     { name = "onnxruntime", specifier = ">=1.20" },

From 5d5d3c7a67046222bc57e5e765313845feeadbc5 Mon Sep 17 00:00:00 2001
From: Cedric Karungu <ckarungu921@gmail.com>
Date: Thu, 8 Oct 2026 23:36:38 +0200
Subject: [PATCH 2/5] fix(local): check the packaged voicing as MP3 and encode
 off the event loop

---
 docs/architecture/studio.md                   |  2 +-
 .../modules/llm/providers/audiocpp/speech.py  |  3 ++-
 .../tests/packaging/test_audiocpp_voicing.py  | 24 +++++++++++++++----
 3 files changed, 22 insertions(+), 7 deletions(-)

diff --git a/docs/architecture/studio.md b/docs/architecture/studio.md
index 1c5bdbb01..ef907dbe9 100644
--- a/docs/architecture/studio.md
+++ b/docs/architecture/studio.md
@@ -82,7 +82,7 @@ Studio's Word and PDF buttons take one of two paths, chosen by [`office/document
 - The job sends the whole spec, fenced, and the instruction in one call, gets the whole revised spec back, and renders it as a draft of that kind would. Nothing is patched or diffed. The rewritten spec is kept only when it renders: a failed refine keeps none, is not retried by Huey, and leaves every earlier version as it was.
 - The agent's `surfsense_read_document` and `surfsense_render_document` refuse any version of a document Studio made, with "This document was made in Studio: refine it there, or make a new document." ([07](../proposals/file-agent/07-create-and-edit-mvp.md), decision 8).
 
-Studio uses no Electron `printToPDF`, which only draws the agent's Word and PowerPoint previews and its sources' pages ([agent](agent.md#previews)), and there is no ffmpeg in `surfsense_local`. A PDF is ReportLab through the builder, a Studio script or a script document. A local podcast is one `audio/mpeg` encoded by LAME, through the `lameenc` package, from each turn's WAV joined; a server's is one `audio/wav` joined from its WAV turns, or one `audio/mpeg` from a server that sends only MP3.
+Studio uses no Electron `printToPDF`, which only draws the agent's Word and PowerPoint previews and its sources' pages ([agent](agent.md#previews)), and there is no ffmpeg in `surfsense_local`. A PDF is ReportLab through the builder, a Studio script or a script document.
 
 ## Script documents
 
diff --git a/surfsense_local/backend/modules/llm/providers/audiocpp/speech.py b/surfsense_local/backend/modules/llm/providers/audiocpp/speech.py
index 05743573a..0ba601200 100644
--- a/surfsense_local/backend/modules/llm/providers/audiocpp/speech.py
+++ b/surfsense_local/backend/modules/llm/providers/audiocpp/speech.py
@@ -109,7 +109,8 @@ class AudioCppSpeech:
                     voiced = await self._voice_each(client, turns, language, speaks)
                 finally:
                     await _unload(client)
-        return encoded_episode(joined_wav(voiced))
+        # Off the event loop: encoding a long episode takes seconds of CPU.
+        return await asyncio.to_thread(lambda: encoded_episode(joined_wav(voiced)))
 
     async def _wait_for_memory(self, seconds: float) -> None:
         deadline = time.monotonic() + seconds
diff --git a/surfsense_local/backend/tests/packaging/test_audiocpp_voicing.py b/surfsense_local/backend/tests/packaging/test_audiocpp_voicing.py
index b0f66e056..89a9c45c7 100644
--- a/surfsense_local/backend/tests/packaging/test_audiocpp_voicing.py
+++ b/surfsense_local/backend/tests/packaging/test_audiocpp_voicing.py
@@ -11,14 +11,12 @@ fails the run: CI sets the folder, and a skip there would pass unseen.
 import asyncio
 import contextlib
 import hashlib
-import io
 import os
 import shutil
 import socket
 import subprocess
 import sys
 import time
-import wave
 from collections.abc import Iterator
 from pathlib import Path
 
@@ -203,8 +201,24 @@ def test_a_curated_model_voices_through_the_app_and_is_given_back(
         sys.stderr.write("\n".join(["audiocpp_server's log, last lines:", *tail]))
         raise
 
-    with wave.open(io.BytesIO(voiced.content)) as joined:
-        assert joined.getframerate() == audio.sample_rate
-        assert joined.getnframes() / joined.getframerate() > 1.0
+    assert voiced.media_type == "audio/mpeg"
+    assert _mp3_rate(voiced.content) == audio.sample_rate
+    # At least a second of speech at 64 kbit/s per channel.
+    assert len(voiced.content) > 64_000 // 8
     listed = httpx.get(f"{server}/v1/models").json()["data"]
     assert not [m["id"] for m in listed if m["loaded"]]
+
+
+# The sample rates an MPEG audio frame header can name, by version.
+_MPEG_RATES = {
+    3: (44100, 48000, 32000),
+    2: (22050, 24000, 16000),
+    0: (11025, 12000, 8000),
+}
+
+
+def _mp3_rate(episode: bytes) -> int:
+    """The sample rate the episode's first MPEG frame header names."""
+    assert episode[0] == 0xFF and episode[1] & 0xE0 == 0xE0, "an MPEG frame first"
+    version = (episode[1] >> 3) & 0b11
+    return _MPEG_RATES[version][(episode[2] >> 2) & 0b11]

From a1ae7c67ac4aae14e1251f379376abf8b4c4b959 Mon Sep 17 00:00:00 2001
From: Cedric Karungu <ckarungu921@gmail.com>
Date: Wed, 14 Oct 2026 00:23:24 +0200
Subject: [PATCH 3/5] fix(local): ship lameenc's and LAME's licences with the
 worker, and encode remote episodes too

The worker bundle now carries bundling/licenses/lameenc/ beside the module:
lameenc's LGPL-3.0 and the GPL-3.0 it builds on (from gnu.org), LAME 3.100's
COPYING and README copyright notice (from its release tarball), and where
lameenc v1.8.4's and LAME 3.100's sources are. The wheel's dist-info holds
lameenc's LGPL only, not LAME's, as the spec and packaging doc had said.

A server that voices in WAV now gets the same MP3 encoding as audio.cpp:
the encoder and the WAV join move to providers/episode_audio/, which both
providers import. The API binary excludes lameenc, which it never needs.
---
 docs/architecture/packaging.md                |   4 +-
 docs/architecture/studio.md                   |   7 +-
 surfsense_local/backend/bundling/api.spec     |   2 +
 .../bundling/licenses/lameenc/GPL-3.0.txt     | 674 ++++++++++++++++++
 .../bundling/licenses/lameenc/LAME-COPYING    | 481 +++++++++++++
 .../bundling/licenses/lameenc/LAME-README     |  44 ++
 .../bundling/licenses/lameenc/SOURCE.txt      |  23 +
 .../licenses/lameenc/lameenc-LGPL-3.0.txt     | 165 +++++
 surfsense_local/backend/bundling/worker.spec  |   9 +-
 .../modules/llm/providers/audiocpp/speech.py  |   4 +-
 .../llm/providers/episode_audio/__init__.py   |   1 +
 .../encoded_episode.py                        |   4 +-
 .../{audiocpp => episode_audio}/joined_wav.py |   0
 .../openai_compatible/joined_replies.py       |   2 +-
 .../llm/providers/openai_compatible/speech.py |   8 +-
 .../unit/bundling/test_lameenc_licences.py    |  32 +
 .../openai_compatible/test_speech.py          |  23 +-
 17 files changed, 1466 insertions(+), 17 deletions(-)
 create mode 100644 surfsense_local/backend/bundling/licenses/lameenc/GPL-3.0.txt
 create mode 100644 surfsense_local/backend/bundling/licenses/lameenc/LAME-COPYING
 create mode 100644 surfsense_local/backend/bundling/licenses/lameenc/LAME-README
 create mode 100644 surfsense_local/backend/bundling/licenses/lameenc/SOURCE.txt
 create mode 100644 surfsense_local/backend/bundling/licenses/lameenc/lameenc-LGPL-3.0.txt
 create mode 100644 surfsense_local/backend/modules/llm/providers/episode_audio/__init__.py
 rename surfsense_local/backend/modules/llm/providers/{audiocpp => episode_audio}/encoded_episode.py (88%)
 rename surfsense_local/backend/modules/llm/providers/{audiocpp => episode_audio}/joined_wav.py (100%)
 create mode 100644 surfsense_local/backend/tests/unit/bundling/test_lameenc_licences.py

diff --git a/docs/architecture/packaging.md b/docs/architecture/packaging.md
index d0497f15d..0cefd641d 100644
--- a/docs/architecture/packaging.md
+++ b/docs/architecture/packaging.md
@@ -41,14 +41,14 @@ What else each spec names, and why the analyser cannot find it on its own:
 | `api.spec` | `collect_submodules("uvicorn")` | uvicorn loads its loop, protocol and lifespan implementations by string |
 | `api.spec` | `onnxruntime` and `tokenizers` libraries | the query encoder's native libraries load from C |
 | `api.spec` | the local model manifest `catalog/local/manifest/models.json`, the remote model manifest `catalog/remote/manifest/models.json`, the chat prompts, the agent's prompt, the agent's `skills/*/SKILL.md` | read by path or through `importlib.resources`; opencode's configuration names the skills folder by path ([agent](agent.md#the-documents-skill)) |
-| `api.spec` | excludes Docling, torch, torchvision, transformers, pandas, scipy and OpenCV | only the worker parses files, and the analyser cannot tell these are optional |
+| `api.spec` | excludes Docling, torch, torchvision, transformers, pandas, scipy, OpenCV and `lameenc` | only the worker parses files and encodes podcasts, and the analyser cannot tell these are optional; the API's voice check keeps the server's WAV |
 | `worker.spec` | the local model manifest, the remote model manifest | Studio finds its chosen image and audio models through the local catalog, and classifies a remote model through the same discovery the API uses; both files are read by path |
 | `worker.spec` | Docling and its packages, RapidOCR, transformers, torchvision | lazy and native imports Docling reaches only on the first PDF |
 | `worker.spec` | python-docx, python-pptx, xlsxwriter, reportlab, openpyxl | the Office formats and document scripts run model-written code that imports them, so no static import exists; openpyxl, named in `pyproject.toml` though Docling brings it too, also reads every document script's workbook for its summary |
 | `worker.spec` | `matplotlib.pyplot`, `matplotlib.backends.backend_pdf`, `matplotlib.backends.backend_svg`, with the matplotlib hook set to the `Agg` backend | document scripts draw charts, and nothing imports matplotlib statically. Naming pyplot runs PyInstaller's hook, which adds `mpl-data` and the backend chosen; a chart saved as PDF or SVG, or `PdfPages`, imports its canvas by name, which the hook does not follow. `Agg` alone keeps Tk's GUI backend out, since scripts run with `MPLBACKEND=Agg` |
 | `worker.spec` | `pandas` | analysis scripts read spreadsheets with it, and only Docling reaches it statically |
 | `worker.spec` | `modules.documents.tasks`, `modules.artifacts.tasks`, `modules.agent.data_analysis.task` | Huey resolves a task by its name |
-| `worker.spec` | `copy_metadata("lameenc")` | a podcast is encoded as MP3 by LAME, statically linked into `lameenc`'s extension, which the analyser finds by its import ([studio](studio.md#voicing-a-podcast)); LAME's LGPL licence text is only in the wheel's dist-info, which ships only when named |
+| `worker.spec` | `copy_metadata("lameenc")`, and `bundling/licenses/lameenc/` as `lameenc-licenses/` | a podcast is encoded as MP3 by LAME, statically linked into `lameenc`'s extension, which the analyser finds by its import ([studio](studio.md#voicing-a-podcast)) and which ships as its own file a user can replace. The wheel's dist-info holds lameenc's LGPL-3.0 text and ships only when named. The folder adds the GPL-3.0 that LGPL-3 §4(b) also asks for, LAME 3.100's `COPYING` and copyright notice from its release tarball, and where lameenc v1.8.4's and LAME 3.100's sources are |
 
 Office support adds nothing to either binary. LibreOffice is downloaded only when the user turns it on, or is the user's own, and runs as its own process from `<data>/runtime/office/` ([office pack](office-pack.md)): the API starts it for a Word or PowerPoint version's previews and a source's pages, the Studio worker for a workbook's recalculation and a PDF conversion. The worker writes recalculated values with lxml, which python-docx already brings; openpyxl reads LibreOffice's values in the worker only, so the API binary needs none.
 
diff --git a/docs/architecture/studio.md b/docs/architecture/studio.md
index 4a04db099..4ef302ac9 100644
--- a/docs/architecture/studio.md
+++ b/docs/architecture/studio.md
@@ -34,7 +34,7 @@ The catalog is a tuple of twelve `Format` rows in [`formats.py`](../../surfsense
 | `mindmap` | Mind map | text_gen | `content/mindmap/` | none; the outline is the body |
 | `flashcards` | Flashcards | text_gen | `content/flashcards/` | deck JSON |
 | `quiz` | Quiz | text_gen | `content/quiz/` | quiz JSON |
-| `podcast` | Podcast | text_gen, audio_gen | `media/audio/podcast/` | MP3 audio, or WAV from a server that sends WAV |
+| `podcast` | Podcast | text_gen, audio_gen | `media/audio/podcast/` | MP3 audio |
 | `image` | Image | image_gen, text_gen | `media/visual/image/` | the image |
 | `infographic` | Infographic | image_gen, text_gen | `media/visual/infographic/` | the image |
 
@@ -179,7 +179,7 @@ It keeps no `spec` and no `version`, so it lists as a document of its own and no
 - **The voices are the model's roster**, from its manifest entry: each voice with the languages it speaks, one for a Kokoro or Kitten voice and all 31 for a Supertonic voice. The brief's language list, A to Z by the name shown in the interface language, and each speaker's voice picker come from it, the picker grouped into female and male voices. The brief opens in American English where a voice speaks it, else in any English, else in the roster's first language, and a remembered brief the chosen model cannot voice falls back to that.
 - **Memory is checked before anything loads, twice.** The worker compares the operating system's available memory (`MemAvailable` on Linux, free plus inactive pages on macOS, `GlobalMemoryStatusEx` on Windows) with the model's peak measured while voicing with every chunk of text full, from its entry, plus 1 GiB, and refuses short of it: "Voicing needs about 3.5 GB free; this computer has 1.8 GB. Supertonic 3 needs about 1.6 GB." The second sentence names the first other curated audio model, in the manifest's order, that would fit, and is left out when none would. It checks first, before the chat model drafts anything, so a machine that cannot voice the episode does not spend minutes writing it. Short there, it gives up the text runtime through the API ([the image job's hold](local-models/admission.md#the-image-jobs-hold)), which lets running replies finish and unloads the chat model, checks again, and lets it go for drafting, because voicing frees that memory anyway: on 16 GB, Qwen3 1.7B resident left 3.3 GB free and refused Kokoro's 3.5 before a word was drafted. At voicing it gives the runtime up again and holds it until the last turn is voiced, so a chat sent meanwhile waits instead of loading the chat model back into memory voicing needs. The router reloads the chat model on the next request after either. Each check waits up to 10 seconds for memory the app is giving back before it refuses, because a starting podcast ends sd-server's idle minutes but Electron stops it only on its next 5-second poll ([the image path](#the-image-path)). audio.cpp's own guard counts only the file it reads.
 - **One speech request per turn**, with the turn's voice. The brief's language goes in the request only for a voice that speaks several; a Kokoro voice's name already fixes its language. The turns are joined with 0.35 s between them.
-- **The episode is MP3**: the joined 16-bit WAV is encoded in-process by LAME through `lameenc`, at 64 kbit/s per channel and the model's own channels and rate, about a sixth of the WAV ([`encoded_episode.py`](../../surfsense_local/backend/modules/llm/providers/audiocpp/encoded_episode.py)). No ffmpeg and nothing on the user's `PATH` is used. Where `lameenc` cannot be imported, as in a checkout that skipped it, the episode stays WAV, which the viewer's `<audio>` plays the same.
+- **The episode is MP3**: the joined 16-bit WAV is encoded in-process by LAME through `lameenc`, at 64 kbit/s per channel and the model's own channels and rate, about a sixth of the WAV ([`encoded_episode.py`](../../surfsense_local/backend/modules/llm/providers/episode_audio/encoded_episode.py)), which a server's WAV episode goes through too. No ffmpeg and nothing on the user's `PATH` is used. Where `lameenc` cannot be imported, as in a checkout that skipped it, the episode stays WAV, which the viewer's `<audio>` plays the same.
 - **A turn the server fails ends the episode with the server's own words**: "audio.cpp could not voice turn 3 of 40: unknown Kokoro voice id: nobody". The server was reached, so this is not "The model could not be reached", which stays for a server that does not answer.
 - **The model is unloaded when voicing ends**, with `POST /v1/tasks/unload_all_models`, success or failure, so its memory is back before the chat model's next job. The server's five-minute idle unload is the backstop.
 
@@ -192,7 +192,7 @@ It keeps no `spec` and no `version`, so it lists as a document of its own and no
 - **The brief reads like the local one.** It adds one line, "Voiced by {model} on {server}. Each line is billed by the provider.", and otherwise keeps the local labels; only its lists differ, the voices one plain list with no gender the server did not state. With no voices yet, the brief shows only a Set up voices button, which opens Settings, and Generate is held. The brief response carries `voices_source` (`local`, `server` or `saved`) and `voiced_by`.
 - **No voice is checked before drafting.** A listed voice came from the server and an added one was heard when it was added, so the brief's own check, that each speaker's voice is one of them, is what stands between the user and a voice the server refuses.
 - **Only OpenAI's `/audio/speech` is spoken.** ElevenLabs, Gemini and other providers with their own speech APIs are reached through a gateway that speaks it for them, OpenRouter or a LiteLLM proxy, rather than through an adapter per provider here.
-- **One `POST /audio/speech` per turn**, asking for `wav`, because only WAV takes a pause between speakers. When the first line's `wav` is refused, the episode asks for `mp3` instead: OpenRouter takes only `mp3` or `pcm`, and its `pcm` states no sample rate. The format is read from the bytes, not from what was asked. WAV turns join as audio.cpp's do, and a reply holding several WAVs back to back is one turn. MP3 turns join frame to frame without a pause, each turn's ID3 tag and Xing frame dropped so a player reads the whole episode's length. Anything else ends the episode with "the server answered, but not as WAV or MP3".
+- **One `POST /audio/speech` per turn**, asking for `wav`, because only WAV takes a pause between speakers. When the first line's `wav` is refused, the episode asks for `mp3` instead: OpenRouter takes only `mp3` or `pcm`, and its `pcm` states no sample rate. The format is read from the bytes, not from what was asked. WAV turns join as audio.cpp's do, and a reply holding several WAVs back to back is one turn. MP3 turns join frame to frame without a pause, each turn's ID3 tag and Xing frame dropped so a player reads the whole episode's length. Anything else ends the episode with "the server answered, but not as WAV or MP3". A WAV episode is then encoded as MP3 as audio.cpp's is; an MP3 one is kept as the server sent it.
 - **Nothing is checked for memory**, since the server spends it.
 - **A failure is not retried.** A refused turn, an unreachable server or an unreadable answer ends the episode with the server's own words, and the job is not retried, because a retry would draft and bill the whole episode again.
 
@@ -322,7 +322,6 @@ Every pipeline returns a `Built`: a `title`, the `markdown` that is always the i
 - A failed script's stored reason repeats the run folder's absolute path in every traceback frame, about 130 characters each against its 500, so a long traceback keeps few of its lines for the agent.
 - Without Office support, a workbook's formulas show the value the script stored with them, since neither the runner nor Studio's viewer recalculates: xlsxwriter stores 0 unless given one, which the skill tells the agent to pass. Excel recalculates when it opens the file. The render result says the values were not recalculated.
 - Only the agent's workbooks are recalculated; Studio's own Spreadsheet format is not.
-- A podcast voiced on a server that sends WAV stays WAV: only audio.cpp's episode goes through the MP3 encoder.
 - A revised copy is cut from its design ([03](../proposals/file-agent/03-editable-artifacts.md)): versions are artifacts, not `artifact_versions` rows, and engines run in the Studio worker's thread, not an engines queue's child process; the agent anchors edits by quote, with no `inspect_document` ids; there are no partial workflows, no per-change accept or reject and no side list of changes, no "With changes (internal)" download and no leak scan, no review record, no reconcile when the original changes, and no "Make a revised copy" in the source viewer. Word headers, footers and footnotes are not edited. An edited workbook is not recalculated until Excel opens it, and a deck has no thumbnails or diff. The report's codes reach the user in English through the agent: the interface does not render them.
 - The Word viewer draws tracked changes and comments with docx-preview's experimental rendering. Its 0.4.0 drops what a fragment holds, which is how it makes a comment's reference, and throws on a comment with no author, so the viewer passes it an element maker that keeps the one and fills the other; a file whose comments it still cannot lay out shows without them. A comment's time shows as Word shows it, unconverted, since Word writes its own clock's time marked UTC. The balloons take about 230 px beside the page, so a file with comments fits the frame at a smaller zoom.
 - The Word viewer fits a file to the frame once, when it loads: dragging the rail's edge afterwards keeps that zoom, where the PDF viewer refits to the page width until it is zoomed by hand ([#2219](https://github.com/MODSetter/SurfSense/issues/2219)).
diff --git a/surfsense_local/backend/bundling/api.spec b/surfsense_local/backend/bundling/api.spec
index af1609e7d..76c984917 100644
--- a/surfsense_local/backend/bundling/api.spec
+++ b/surfsense_local/backend/bundling/api.spec
@@ -98,6 +98,8 @@ a = Analysis(
         "onnxruntime.transformers",
         "onnxruntime.quantization",
         "onnxruntime.tools",
+        # Only the worker encodes a podcast; the API's voice check keeps its WAV.
+        "lameenc",
     ],
 )
 pyz = PYZ(a.pure)
diff --git a/surfsense_local/backend/bundling/licenses/lameenc/GPL-3.0.txt b/surfsense_local/backend/bundling/licenses/lameenc/GPL-3.0.txt
new file mode 100644
index 000000000..f288702d2
--- /dev/null
+++ b/surfsense_local/backend/bundling/licenses/lameenc/GPL-3.0.txt
@@ -0,0 +1,674 @@
+                    GNU GENERAL PUBLIC LICENSE
+                       Version 3, 29 June 2007
+
+ Copyright (C) 2007 Free Software Foundation, Inc. <https://fsf.org/>
+ Everyone is permitted to copy and distribute verbatim copies
+ of this license document, but changing it is not allowed.
+
+                            Preamble
+
+  The GNU General Public License is a free, copyleft license for
+software and other kinds of works.
+
+  The licenses for most software and other practical works are designed
+to take away your freedom to share and change the works.  By contrast,
+the GNU General Public License is intended to guarantee your freedom to
+share and change all versions of a program--to make sure it remains free
+software for all its users.  We, the Free Software Foundation, use the
+GNU General Public License for most of our software; it applies also to
+any other work released this way by its authors.  You can apply it to
+your programs, too.
+
+  When we speak of free software, we are referring to freedom, not
+price.  Our General Public Licenses are designed to make sure that you
+have the freedom to distribute copies of free software (and charge for
+them if you wish), that you receive source code or can get it if you
+want it, that you can change the software or use pieces of it in new
+free programs, and that you know you can do these things.
+
+  To protect your rights, we need to prevent others from denying you
+these rights or asking you to surrender the rights.  Therefore, you have
+certain responsibilities if you distribute copies of the software, or if
+you modify it: responsibilities to respect the freedom of others.
+
+  For example, if you distribute copies of such a program, whether
+gratis or for a fee, you must pass on to the recipients the same
+freedoms that you received.  You must make sure that they, too, receive
+or can get the source code.  And you must show them these terms so they
+know their rights.
+
+  Developers that use the GNU GPL protect your rights with two steps:
+(1) assert copyright on the software, and (2) offer you this License
+giving you legal permission to copy, distribute and/or modify it.
+
+  For the developers' and authors' protection, the GPL clearly explains
+that there is no warranty for this free software.  For both users' and
+authors' sake, the GPL requires that modified versions be marked as
+changed, so that their problems will not be attributed erroneously to
+authors of previous versions.
+
+  Some devices are designed to deny users access to install or run
+modified versions of the software inside them, although the manufacturer
+can do so.  This is fundamentally incompatible with the aim of
+protecting users' freedom to change the software.  The systematic
+pattern of such abuse occurs in the area of products for individuals to
+use, which is precisely where it is most unacceptable.  Therefore, we
+have designed this version of the GPL to prohibit the practice for those
+products.  If such problems arise substantially in other domains, we
+stand ready to extend this provision to those domains in future versions
+of the GPL, as needed to protect the freedom of users.
+
+  Finally, every program is threatened constantly by software patents.
+States should not allow patents to restrict development and use of
+software on general-purpose computers, but in those that do, we wish to
+avoid the special danger that patents applied to a free program could
+make it effectively proprietary.  To prevent this, the GPL assures that
+patents cannot be used to render the program non-free.
+
+  The precise terms and conditions for copying, distribution and
+modification follow.
+
+                       TERMS AND CONDITIONS
+
+  0. Definitions.
+
+  "This License" refers to version 3 of the GNU General Public License.
+
+  "Copyright" also means copyright-like laws that apply to other kinds of
+works, such as semiconductor masks.
+
+  "The Program" refers to any copyrightable work licensed under this
+License.  Each licensee is addressed as "you".  "Licensees" and
+"recipients" may be individuals or organizations.
+
+  To "modify" a work means to copy from or adapt all or part of the work
+in a fashion requiring copyright permission, other than the making of an
+exact copy.  The resulting work is called a "modified version" of the
+earlier work or a work "based on" the earlier work.
+
+  A "covered work" means either the unmodified Program or a work based
+on the Program.
+
+  To "propagate" a work means to do anything with it that, without
+permission, would make you directly or secondarily liable for
+infringement under applicable copyright law, except executing it on a
+computer or modifying a private copy.  Propagation includes copying,
+distribution (with or without modification), making available to the
+public, and in some countries other activities as well.
+
+  To "convey" a work means any kind of propagation that enables other
+parties to make or receive copies.  Mere interaction with a user through
+a computer network, with no transfer of a copy, is not conveying.
+
+  An interactive user interface displays "Appropriate Legal Notices"
+to the extent that it includes a convenient and prominently visible
+feature that (1) displays an appropriate copyright notice, and (2)
+tells the user that there is no warranty for the work (except to the
+extent that warranties are provided), that licensees may convey the
+work under this License, and how to view a copy of this License.  If
+the interface presents a list of user commands or options, such as a
+menu, a prominent item in the list meets this criterion.
+
+  1. Source Code.
+
+  The "source code" for a work means the preferred form of the work
+for making modifications to it.  "Object code" means any non-source
+form of a work.
+
+  A "Standard Interface" means an interface that either is an official
+standard defined by a recognized standards body, or, in the case of
+interfaces specified for a particular programming language, one that
+is widely used among developers working in that language.
+
+  The "System Libraries" of an executable work include anything, other
+than the work as a whole, that (a) is included in the normal form of
+packaging a Major Component, but which is not part of that Major
+Component, and (b) serves only to enable use of the work with that
+Major Component, or to implement a Standard Interface for which an
+implementation is available to the public in source code form.  A
+"Major Component", in this context, means a major essential component
+(kernel, window system, and so on) of the specific operating system
+(if any) on which the executable work runs, or a compiler used to
+produce the work, or an object code interpreter used to run it.
+
+  The "Corresponding Source" for a work in object code form means all
+the source code needed to generate, install, and (for an executable
+work) run the object code and to modify the work, including scripts to
+control those activities.  However, it does not include the work's
+System Libraries, or general-purpose tools or generally available free
+programs which are used unmodified in performing those activities but
+which are not part of the work.  For example, Corresponding Source
+includes interface definition files associated with source files for
+the work, and the source code for shared libraries and dynamically
+linked subprograms that the work is specifically designed to require,
+such as by intimate data communication or control flow between those
+subprograms and other parts of the work.
+
+  The Corresponding Source need not include anything that users
+can regenerate automatically from other parts of the Corresponding
+Source.
+
+  The Corresponding Source for a work in source code form is that
+same work.
+
+  2. Basic Permissions.
+
+  All rights granted under this License are granted for the term of
+copyright on the Program, and are irrevocable provided the stated
+conditions are met.  This License explicitly affirms your unlimited
+permission to run the unmodified Program.  The output from running a
+covered work is covered by this License only if the output, given its
+content, constitutes a covered work.  This License acknowledges your
+rights of fair use or other equivalent, as provided by copyright law.
+
+  You may make, run and propagate covered works that you do not
+convey, without conditions so long as your license otherwise remains
+in force.  You may convey covered works to others for the sole purpose
+of having them make modifications exclusively for you, or provide you
+with facilities for running those works, provided that you comply with
+the terms of this License in conveying all material for which you do
+not control copyright.  Those thus making or running the covered works
+for you must do so exclusively on your behalf, under your direction
+and control, on terms that prohibit them from making any copies of
+your copyrighted material outside their relationship with you.
+
+  Conveying under any other circumstances is permitted solely under
+the conditions stated below.  Sublicensing is not allowed; section 10
+makes it unnecessary.
+
+  3. Protecting Users' Legal Rights From Anti-Circumvention Law.
+
+  No covered work shall be deemed part of an effective technological
+measure under any applicable law fulfilling obligations under article
+11 of the WIPO copyright treaty adopted on 20 December 1996, or
+similar laws prohibiting or restricting circumvention of such
+measures.
+
+  When you convey a covered work, you waive any legal power to forbid
+circumvention of technological measures to the extent such circumvention
+is effected by exercising rights under this License with respect to
+the covered work, and you disclaim any intention to limit operation or
+modification of the work as a means of enforcing, against the work's
+users, your or third parties' legal rights to forbid circumvention of
+technological measures.
+
+  4. Conveying Verbatim Copies.
+
+  You may convey verbatim copies of the Program's source code as you
+receive it, in any medium, provided that you conspicuously and
+appropriately publish on each copy an appropriate copyright notice;
+keep intact all notices stating that this License and any
+non-permissive terms added in accord with section 7 apply to the code;
+keep intact all notices of the absence of any warranty; and give all
+recipients a copy of this License along with the Program.
+
+  You may charge any price or no price for each copy that you convey,
+and you may offer support or warranty protection for a fee.
+
+  5. Conveying Modified Source Versions.
+
+  You may convey a work based on the Program, or the modifications to
+produce it from the Program, in the form of source code under the
+terms of section 4, provided that you also meet all of these conditions:
+
+    a) The work must carry prominent notices stating that you modified
+    it, and giving a relevant date.
+
+    b) The work must carry prominent notices stating that it is
+    released under this License and any conditions added under section
+    7.  This requirement modifies the requirement in section 4 to
+    "keep intact all notices".
+
+    c) You must license the entire work, as a whole, under this
+    License to anyone who comes into possession of a copy.  This
+    License will therefore apply, along with any applicable section 7
+    additional terms, to the whole of the work, and all its parts,
+    regardless of how they are packaged.  This License gives no
+    permission to license the work in any other way, but it does not
+    invalidate such permission if you have separately received it.
+
+    d) If the work has interactive user interfaces, each must display
+    Appropriate Legal Notices; however, if the Program has interactive
+    interfaces that do not display Appropriate Legal Notices, your
+    work need not make them do so.
+
+  A compilation of a covered work with other separate and independent
+works, which are not by their nature extensions of the covered work,
+and which are not combined with it such as to form a larger program,
+in or on a volume of a storage or distribution medium, is called an
+"aggregate" if the compilation and its resulting copyright are not
+used to limit the access or legal rights of the compilation's users
+beyond what the individual works permit.  Inclusion of a covered work
+in an aggregate does not cause this License to apply to the other
+parts of the aggregate.
+
+  6. Conveying Non-Source Forms.
+
+  You may convey a covered work in object code form under the terms
+of sections 4 and 5, provided that you also convey the
+machine-readable Corresponding Source under the terms of this License,
+in one of these ways:
+
+    a) Convey the object code in, or embodied in, a physical product
+    (including a physical distribution medium), accompanied by the
+    Corresponding Source fixed on a durable physical medium
+    customarily used for software interchange.
+
+    b) Convey the object code in, or embodied in, a physical product
+    (including a physical distribution medium), accompanied by a
+    written offer, valid for at least three years and valid for as
+    long as you offer spare parts or customer support for that product
+    model, to give anyone who possesses the object code either (1) a
+    copy of the Corresponding Source for all the software in the
+    product that is covered by this License, on a durable physical
+    medium customarily used for software interchange, for a price no
+    more than your reasonable cost of physically performing this
+    conveying of source, or (2) access to copy the
+    Corresponding Source from a network server at no charge.
+
+    c) Convey individual copies of the object code with a copy of the
+    written offer to provide the Corresponding Source.  This
+    alternative is allowed only occasionally and noncommercially, and
+    only if you received the object code with such an offer, in accord
+    with subsection 6b.
+
+    d) Convey the object code by offering access from a designated
+    place (gratis or for a charge), and offer equivalent access to the
+    Corresponding Source in the same way through the same place at no
+    further charge.  You need not require recipients to copy the
+    Corresponding Source along with the object code.  If the place to
+    copy the object code is a network server, the Corresponding Source
+    may be on a different server (operated by you or a third party)
+    that supports equivalent copying facilities, provided you maintain
+    clear directions next to the object code saying where to find the
+    Corresponding Source.  Regardless of what server hosts the
+    Corresponding Source, you remain obligated to ensure that it is
+    available for as long as needed to satisfy these requirements.
+
+    e) Convey the object code using peer-to-peer transmission, provided
+    you inform other peers where the object code and Corresponding
+    Source of the work are being offered to the general public at no
+    charge under subsection 6d.
+
+  A separable portion of the object code, whose source code is excluded
+from the Corresponding Source as a System Library, need not be
+included in conveying the object code work.
+
+  A "User Product" is either (1) a "consumer product", which means any
+tangible personal property which is normally used for personal, family,
+or household purposes, or (2) anything designed or sold for incorporation
+into a dwelling.  In determining whether a product is a consumer product,
+doubtful cases shall be resolved in favor of coverage.  For a particular
+product received by a particular user, "normally used" refers to a
+typical or common use of that class of product, regardless of the status
+of the particular user or of the way in which the particular user
+actually uses, or expects or is expected to use, the product.  A product
+is a consumer product regardless of whether the product has substantial
+commercial, industrial or non-consumer uses, unless such uses represent
+the only significant mode of use of the product.
+
+  "Installation Information" for a User Product means any methods,
+procedures, authorization keys, or other information required to install
+and execute modified versions of a covered work in that User Product from
+a modified version of its Corresponding Source.  The information must
+suffice to ensure that the continued functioning of the modified object
+code is in no case prevented or interfered with solely because
+modification has been made.
+
+  If you convey an object code work under this section in, or with, or
+specifically for use in, a User Product, and the conveying occurs as
+part of a transaction in which the right of possession and use of the
+User Product is transferred to the recipient in perpetuity or for a
+fixed term (regardless of how the transaction is characterized), the
+Corresponding Source conveyed under this section must be accompanied
+by the Installation Information.  But this requirement does not apply
+if neither you nor any third party retains the ability to install
+modified object code on the User Product (for example, the work has
+been installed in ROM).
+
+  The requirement to provide Installation Information does not include a
+requirement to continue to provide support service, warranty, or updates
+for a work that has been modified or installed by the recipient, or for
+the User Product in which it has been modified or installed.  Access to a
+network may be denied when the modification itself materially and
+adversely affects the operation of the network or violates the rules and
+protocols for communication across the network.
+
+  Corresponding Source conveyed, and Installation Information provided,
+in accord with this section must be in a format that is publicly
+documented (and with an implementation available to the public in
+source code form), and must require no special password or key for
+unpacking, reading or copying.
+
+  7. Additional Terms.
+
+  "Additional permissions" are terms that supplement the terms of this
+License by making exceptions from one or more of its conditions.
+Additional permissions that are applicable to the entire Program shall
+be treated as though they were included in this License, to the extent
+that they are valid under applicable law.  If additional permissions
+apply only to part of the Program, that part may be used separately
+under those permissions, but the entire Program remains governed by
+this License without regard to the additional permissions.
+
+  When you convey a copy of a covered work, you may at your option
+remove any additional permissions from that copy, or from any part of
+it.  (Additional permissions may be written to require their own
+removal in certain cases when you modify the work.)  You may place
+additional permissions on material, added by you to a covered work,
+for which you have or can give appropriate copyright permission.
+
+  Notwithstanding any other provision of this License, for material you
+add to a covered work, you may (if authorized by the copyright holders of
+that material) supplement the terms of this License with terms:
+
+    a) Disclaiming warranty or limiting liability differently from the
+    terms of sections 15 and 16 of this License; or
+
+    b) Requiring preservation of specified reasonable legal notices or
+    author attributions in that material or in the Appropriate Legal
+    Notices displayed by works containing it; or
+
+    c) Prohibiting misrepresentation of the origin of that material, or
+    requiring that modified versions of such material be marked in
+    reasonable ways as different from the original version; or
+
+    d) Limiting the use for publicity purposes of names of licensors or
+    authors of the material; or
+
+    e) Declining to grant rights under trademark law for use of some
+    trade names, trademarks, or service marks; or
+
+    f) Requiring indemnification of licensors and authors of that
+    material by anyone who conveys the material (or modified versions of
+    it) with contractual assumptions of liability to the recipient, for
+    any liability that these contractual assumptions directly impose on
+    those licensors and authors.
+
+  All other non-permissive additional terms are considered "further
+restrictions" within the meaning of section 10.  If the Program as you
+received it, or any part of it, contains a notice stating that it is
+governed by this License along with a term that is a further
+restriction, you may remove that term.  If a license document contains
+a further restriction but permits relicensing or conveying under this
+License, you may add to a covered work material governed by the terms
+of that license document, provided that the further restriction does
+not survive such relicensing or conveying.
+
+  If you add terms to a covered work in accord with this section, you
+must place, in the relevant source files, a statement of the
+additional terms that apply to those files, or a notice indicating
+where to find the applicable terms.
+
+  Additional terms, permissive or non-permissive, may be stated in the
+form of a separately written license, or stated as exceptions;
+the above requirements apply either way.
+
+  8. Termination.
+
+  You may not propagate or modify a covered work except as expressly
+provided under this License.  Any attempt otherwise to propagate or
+modify it is void, and will automatically terminate your rights under
+this License (including any patent licenses granted under the third
+paragraph of section 11).
+
+  However, if you cease all violation of this License, then your
+license from a particular copyright holder is reinstated (a)
+provisionally, unless and until the copyright holder explicitly and
+finally terminates your license, and (b) permanently, if the copyright
+holder fails to notify you of the violation by some reasonable means
+prior to 60 days after the cessation.
+
+  Moreover, your license from a particular copyright holder is
+reinstated permanently if the copyright holder notifies you of the
+violation by some reasonable means, this is the first time you have
+received notice of violation of this License (for any work) from that
+copyright holder, and you cure the violation prior to 30 days after
+your receipt of the notice.
+
+  Termination of your rights under this section does not terminate the
+licenses of parties who have received copies or rights from you under
+this License.  If your rights have been terminated and not permanently
+reinstated, you do not qualify to receive new licenses for the same
+material under section 10.
+
+  9. Acceptance Not Required for Having Copies.
+
+  You are not required to accept this License in order to receive or
+run a copy of the Program.  Ancillary propagation of a covered work
+occurring solely as a consequence of using peer-to-peer transmission
+to receive a copy likewise does not require acceptance.  However,
+nothing other than this License grants you permission to propagate or
+modify any covered work.  These actions infringe copyright if you do
+not accept this License.  Therefore, by modifying or propagating a
+covered work, you indicate your acceptance of this License to do so.
+
+  10. Automatic Licensing of Downstream Recipients.
+
+  Each time you convey a covered work, the recipient automatically
+receives a license from the original licensors, to run, modify and
+propagate that work, subject to this License.  You are not responsible
+for enforcing compliance by third parties with this License.
+
+  An "entity transaction" is a transaction transferring control of an
+organization, or substantially all assets of one, or subdividing an
+organization, or merging organizations.  If propagation of a covered
+work results from an entity transaction, each party to that
+transaction who receives a copy of the work also receives whatever
+licenses to the work the party's predecessor in interest had or could
+give under the previous paragraph, plus a right to possession of the
+Corresponding Source of the work from the predecessor in interest, if
+the predecessor has it or can get it with reasonable efforts.
+
+  You may not impose any further restrictions on the exercise of the
+rights granted or affirmed under this License.  For example, you may
+not impose a license fee, royalty, or other charge for exercise of
+rights granted under this License, and you may not initiate litigation
+(including a cross-claim or counterclaim in a lawsuit) alleging that
+any patent claim is infringed by making, using, selling, offering for
+sale, or importing the Program or any portion of it.
+
+  11. Patents.
+
+  A "contributor" is a copyright holder who authorizes use under this
+License of the Program or a work on which the Program is based.  The
+work thus licensed is called the contributor's "contributor version".
+
+  A contributor's "essential patent claims" are all patent claims
+owned or controlled by the contributor, whether already acquired or
+hereafter acquired, that would be infringed by some manner, permitted
+by this License, of making, using, or selling its contributor version,
+but do not include claims that would be infringed only as a
+consequence of further modification of the contributor version.  For
+purposes of this definition, "control" includes the right to grant
+patent sublicenses in a manner consistent with the requirements of
+this License.
+
+  Each contributor grants you a non-exclusive, worldwide, royalty-free
+patent license under the contributor's essential patent claims, to
+make, use, sell, offer for sale, import and otherwise run, modify and
+propagate the contents of its contributor version.
+
+  In the following three paragraphs, a "patent license" is any express
+agreement or commitment, however denominated, not to enforce a patent
+(such as an express permission to practice a patent or covenant not to
+sue for patent infringement).  To "grant" such a patent license to a
+party means to make such an agreement or commitment not to enforce a
+patent against the party.
+
+  If you convey a covered work, knowingly relying on a patent license,
+and the Corresponding Source of the work is not available for anyone
+to copy, free of charge and under the terms of this License, through a
+publicly available network server or other readily accessible means,
+then you must either (1) cause the Corresponding Source to be so
+available, or (2) arrange to deprive yourself of the benefit of the
+patent license for this particular work, or (3) arrange, in a manner
+consistent with the requirements of this License, to extend the patent
+license to downstream recipients.  "Knowingly relying" means you have
+actual knowledge that, but for the patent license, your conveying the
+covered work in a country, or your recipient's use of the covered work
+in a country, would infringe one or more identifiable patents in that
+country that you have reason to believe are valid.
+
+  If, pursuant to or in connection with a single transaction or
+arrangement, you convey, or propagate by procuring conveyance of, a
+covered work, and grant a patent license to some of the parties
+receiving the covered work authorizing them to use, propagate, modify
+or convey a specific copy of the covered work, then the patent license
+you grant is automatically extended to all recipients of the covered
+work and works based on it.
+
+  A patent license is "discriminatory" if it does not include within
+the scope of its coverage, prohibits the exercise of, or is
+conditioned on the non-exercise of one or more of the rights that are
+specifically granted under this License.  You may not convey a covered
+work if you are a party to an arrangement with a third party that is
+in the business of distributing software, under which you make payment
+to the third party based on the extent of your activity of conveying
+the work, and under which the third party grants, to any of the
+parties who would receive the covered work from you, a discriminatory
+patent license (a) in connection with copies of the covered work
+conveyed by you (or copies made from those copies), or (b) primarily
+for and in connection with specific products or compilations that
+contain the covered work, unless you entered into that arrangement,
+or that patent license was granted, prior to 28 March 2007.
+
+  Nothing in this License shall be construed as excluding or limiting
+any implied license or other defenses to infringement that may
+otherwise be available to you under applicable patent law.
+
+  12. No Surrender of Others' Freedom.
+
+  If conditions are imposed on you (whether by court order, agreement or
+otherwise) that contradict the conditions of this License, they do not
+excuse you from the conditions of this License.  If you cannot convey a
+covered work so as to satisfy simultaneously your obligations under this
+License and any other pertinent obligations, then as a consequence you may
+not convey it at all.  For example, if you agree to terms that obligate you
+to collect a royalty for further conveying from those to whom you convey
+the Program, the only way you could satisfy both those terms and this
+License would be to refrain entirely from conveying the Program.
+
+  13. Use with the GNU Affero General Public License.
+
+  Notwithstanding any other provision of this License, you have
+permission to link or combine any covered work with a work licensed
+under version 3 of the GNU Affero General Public License into a single
+combined work, and to convey the resulting work.  The terms of this
+License will continue to apply to the part which is the covered work,
+but the special requirements of the GNU Affero General Public License,
+section 13, concerning interaction through a network will apply to the
+combination as such.
+
+  14. Revised Versions of this License.
+
+  The Free Software Foundation may publish revised and/or new versions of
+the GNU General Public License from time to time.  Such new versions will
+be similar in spirit to the present version, but may differ in detail to
+address new problems or concerns.
+
+  Each version is given a distinguishing version number.  If the
+Program specifies that a certain numbered version of the GNU General
+Public License "or any later version" applies to it, you have the
+option of following the terms and conditions either of that numbered
+version or of any later version published by the Free Software
+Foundation.  If the Program does not specify a version number of the
+GNU General Public License, you may choose any version ever published
+by the Free Software Foundation.
+
+  If the Program specifies that a proxy can decide which future
+versions of the GNU General Public License can be used, that proxy's
+public statement of acceptance of a version permanently authorizes you
+to choose that version for the Program.
+
+  Later license versions may give you additional or different
+permissions.  However, no additional obligations are imposed on any
+author or copyright holder as a result of your choosing to follow a
+later version.
+
+  15. Disclaimer of Warranty.
+
+  THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY
+APPLICABLE LAW.  EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT
+HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY
+OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO,
+THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+PURPOSE.  THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM
+IS WITH YOU.  SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF
+ALL NECESSARY SERVICING, REPAIR OR CORRECTION.
+
+  16. Limitation of Liability.
+
+  IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING
+WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS
+THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY
+GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE
+USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF
+DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD
+PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS),
+EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF
+SUCH DAMAGES.
+
+  17. Interpretation of Sections 15 and 16.
+
+  If the disclaimer of warranty and limitation of liability provided
+above cannot be given local legal effect according to their terms,
+reviewing courts shall apply local law that most closely approximates
+an absolute waiver of all civil liability in connection with the
+Program, unless a warranty or assumption of liability accompanies a
+copy of the Program in return for a fee.
+
+                     END OF TERMS AND CONDITIONS
+
+            How to Apply These Terms to Your New Programs
+
+  If you develop a new program, and you want it to be of the greatest
+possible use to the public, the best way to achieve this is to make it
+free software which everyone can redistribute and change under these terms.
+
+  To do so, attach the following notices to the program.  It is safest
+to attach them to the start of each source file to most effectively
+state the exclusion of warranty; and each file should have at least
+the "copyright" line and a pointer to where the full notice is found.
+
+    <one line to give the program's name and a brief idea of what it does.>
+    Copyright (C) <year>  <name of author>
+
+    This program is free software: you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation, either version 3 of the License, or
+    (at your option) any later version.
+
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU General Public License for more details.
+
+    You should have received a copy of the GNU General Public License
+    along with this program.  If not, see <https://www.gnu.org/licenses/>.
+
+Also add information on how to contact you by electronic and paper mail.
+
+  If the program does terminal interaction, make it output a short
+notice like this when it starts in an interactive mode:
+
+    <program>  Copyright (C) <year>  <name of author>
+    This program comes with ABSOLUTELY NO WARRANTY; for details type `show w'.
+    This is free software, and you are welcome to redistribute it
+    under certain conditions; type `show c' for details.
+
+The hypothetical commands `show w' and `show c' should show the appropriate
+parts of the General Public License.  Of course, your program's commands
+might be different; for a GUI interface, you would use an "about box".
+
+  You should also get your employer (if you work as a programmer) or school,
+if any, to sign a "copyright disclaimer" for the program, if necessary.
+For more information on this, and how to apply and follow the GNU GPL, see
+<https://www.gnu.org/licenses/>.
+
+  The GNU General Public License does not permit incorporating your program
+into proprietary programs.  If your program is a subroutine library, you
+may consider it more useful to permit linking proprietary applications with
+the library.  If this is what you want to do, use the GNU Lesser General
+Public License instead of this License.  But first, please read
+<https://www.gnu.org/licenses/why-not-lgpl.html>.
diff --git a/surfsense_local/backend/bundling/licenses/lameenc/LAME-COPYING b/surfsense_local/backend/bundling/licenses/lameenc/LAME-COPYING
new file mode 100644
index 000000000..f5030495b
--- /dev/null
+++ b/surfsense_local/backend/bundling/licenses/lameenc/LAME-COPYING
@@ -0,0 +1,481 @@
+		  GNU LIBRARY GENERAL PUBLIC LICENSE
+		       Version 2, June 1991
+
+ Copyright (C) 1991 Free Software Foundation, Inc.
+    		    59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ Everyone is permitted to copy and distribute verbatim copies
+ of this license document, but changing it is not allowed.
+
+[This is the first released version of the library GPL.  It is
+ numbered 2 because it goes with version 2 of the ordinary GPL.]
+
+			    Preamble
+
+  The licenses for most software are designed to take away your
+freedom to share and change it.  By contrast, the GNU General Public
+Licenses are intended to guarantee your freedom to share and change
+free software--to make sure the software is free for all its users.
+
+  This license, the Library General Public License, applies to some
+specially designated Free Software Foundation software, and to any
+other libraries whose authors decide to use it.  You can use it for
+your libraries, too.
+
+  When we speak of free software, we are referring to freedom, not
+price.  Our General Public Licenses are designed to make sure that you
+have the freedom to distribute copies of free software (and charge for
+this service if you wish), that you receive source code or can get it
+if you want it, that you can change the software or use pieces of it
+in new free programs; and that you know you can do these things.
+
+  To protect your rights, we need to make restrictions that forbid
+anyone to deny you these rights or to ask you to surrender the rights.
+These restrictions translate to certain responsibilities for you if
+you distribute copies of the library, or if you modify it.
+
+  For example, if you distribute copies of the library, whether gratis
+or for a fee, you must give the recipients all the rights that we gave
+you.  You must make sure that they, too, receive or can get the source
+code.  If you link a program with the library, you must provide
+complete object files to the recipients so that they can relink them
+with the library, after making changes to the library and recompiling
+it.  And you must show them these terms so they know their rights.
+
+  Our method of protecting your rights has two steps: (1) copyright
+the library, and (2) offer you this license which gives you legal
+permission to copy, distribute and/or modify the library.
+
+  Also, for each distributor's protection, we want to make certain
+that everyone understands that there is no warranty for this free
+library.  If the library is modified by someone else and passed on, we
+want its recipients to know that what they have is not the original
+version, so that any problems introduced by others will not reflect on
+the original authors' reputations.
+
+  Finally, any free program is threatened constantly by software
+patents.  We wish to avoid the danger that companies distributing free
+software will individually obtain patent licenses, thus in effect
+transforming the program into proprietary software.  To prevent this,
+we have made it clear that any patent must be licensed for everyone's
+free use or not licensed at all.
+
+  Most GNU software, including some libraries, is covered by the ordinary
+GNU General Public License, which was designed for utility programs.  This
+license, the GNU Library General Public License, applies to certain
+designated libraries.  This license is quite different from the ordinary
+one; be sure to read it in full, and don't assume that anything in it is
+the same as in the ordinary license.
+
+  The reason we have a separate public license for some libraries is that
+they blur the distinction we usually make between modifying or adding to a
+program and simply using it.  Linking a program with a library, without
+changing the library, is in some sense simply using the library, and is
+analogous to running a utility program or application program.  However, in
+a textual and legal sense, the linked executable is a combined work, a
+derivative of the original library, and the ordinary General Public License
+treats it as such.
+
+  Because of this blurred distinction, using the ordinary General
+Public License for libraries did not effectively promote software
+sharing, because most developers did not use the libraries.  We
+concluded that weaker conditions might promote sharing better.
+
+  However, unrestricted linking of non-free programs would deprive the
+users of those programs of all benefit from the free status of the
+libraries themselves.  This Library General Public License is intended to
+permit developers of non-free programs to use free libraries, while
+preserving your freedom as a user of such programs to change the free
+libraries that are incorporated in them.  (We have not seen how to achieve
+this as regards changes in header files, but we have achieved it as regards
+changes in the actual functions of the Library.)  The hope is that this
+will lead to faster development of free libraries.
+
+  The precise terms and conditions for copying, distribution and
+modification follow.  Pay close attention to the difference between a
+"work based on the library" and a "work that uses the library".  The
+former contains code derived from the library, while the latter only
+works together with the library.
+
+  Note that it is possible for a library to be covered by the ordinary
+General Public License rather than by this special one.
+
+		  GNU LIBRARY GENERAL PUBLIC LICENSE
+   TERMS AND CONDITIONS FOR COPYING, DISTRIBUTION AND MODIFICATION
+
+  0. This License Agreement applies to any software library which
+contains a notice placed by the copyright holder or other authorized
+party saying it may be distributed under the terms of this Library
+General Public License (also called "this License").  Each licensee is
+addressed as "you".
+
+  A "library" means a collection of software functions and/or data
+prepared so as to be conveniently linked with application programs
+(which use some of those functions and data) to form executables.
+
+  The "Library", below, refers to any such software library or work
+which has been distributed under these terms.  A "work based on the
+Library" means either the Library or any derivative work under
+copyright law: that is to say, a work containing the Library or a
+portion of it, either verbatim or with modifications and/or translated
+straightforwardly into another language.  (Hereinafter, translation is
+included without limitation in the term "modification".)
+
+  "Source code" for a work means the preferred form of the work for
+making modifications to it.  For a library, complete source code means
+all the source code for all modules it contains, plus any associated
+interface definition files, plus the scripts used to control compilation
+and installation of the library.
+
+  Activities other than copying, distribution and modification are not
+covered by this License; they are outside its scope.  The act of
+running a program using the Library is not restricted, and output from
+such a program is covered only if its contents constitute a work based
+on the Library (independent of the use of the Library in a tool for
+writing it).  Whether that is true depends on what the Library does
+and what the program that uses the Library does.
+  
+  1. You may copy and distribute verbatim copies of the Library's
+complete source code as you receive it, in any medium, provided that
+you conspicuously and appropriately publish on each copy an
+appropriate copyright notice and disclaimer of warranty; keep intact
+all the notices that refer to this License and to the absence of any
+warranty; and distribute a copy of this License along with the
+Library.
+
+  You may charge a fee for the physical act of transferring a copy,
+and you may at your option offer warranty protection in exchange for a
+fee.
+
+  2. You may modify your copy or copies of the Library or any portion
+of it, thus forming a work based on the Library, and copy and
+distribute such modifications or work under the terms of Section 1
+above, provided that you also meet all of these conditions:
+
+    a) The modified work must itself be a software library.
+
+    b) You must cause the files modified to carry prominent notices
+    stating that you changed the files and the date of any change.
+
+    c) You must cause the whole of the work to be licensed at no
+    charge to all third parties under the terms of this License.
+
+    d) If a facility in the modified Library refers to a function or a
+    table of data to be supplied by an application program that uses
+    the facility, other than as an argument passed when the facility
+    is invoked, then you must make a good faith effort to ensure that,
+    in the event an application does not supply such function or
+    table, the facility still operates, and performs whatever part of
+    its purpose remains meaningful.
+
+    (For example, a function in a library to compute square roots has
+    a purpose that is entirely well-defined independent of the
+    application.  Therefore, Subsection 2d requires that any
+    application-supplied function or table used by this function must
+    be optional: if the application does not supply it, the square
+    root function must still compute square roots.)
+
+These requirements apply to the modified work as a whole.  If
+identifiable sections of that work are not derived from the Library,
+and can be reasonably considered independent and separate works in
+themselves, then this License, and its terms, do not apply to those
+sections when you distribute them as separate works.  But when you
+distribute the same sections as part of a whole which is a work based
+on the Library, the distribution of the whole must be on the terms of
+this License, whose permissions for other licensees extend to the
+entire whole, and thus to each and every part regardless of who wrote
+it.
+
+Thus, it is not the intent of this section to claim rights or contest
+your rights to work written entirely by you; rather, the intent is to
+exercise the right to control the distribution of derivative or
+collective works based on the Library.
+
+In addition, mere aggregation of another work not based on the Library
+with the Library (or with a work based on the Library) on a volume of
+a storage or distribution medium does not bring the other work under
+the scope of this License.
+
+  3. You may opt to apply the terms of the ordinary GNU General Public
+License instead of this License to a given copy of the Library.  To do
+this, you must alter all the notices that refer to this License, so
+that they refer to the ordinary GNU General Public License, version 2,
+instead of to this License.  (If a newer version than version 2 of the
+ordinary GNU General Public License has appeared, then you can specify
+that version instead if you wish.)  Do not make any other change in
+these notices.
+
+  Once this change is made in a given copy, it is irreversible for
+that copy, so the ordinary GNU General Public License applies to all
+subsequent copies and derivative works made from that copy.
+
+  This option is useful when you wish to copy part of the code of
+the Library into a program that is not a library.
+
+  4. You may copy and distribute the Library (or a portion or
+derivative of it, under Section 2) in object code or executable form
+under the terms of Sections 1 and 2 above provided that you accompany
+it with the complete corresponding machine-readable source code, which
+must be distributed under the terms of Sections 1 and 2 above on a
+medium customarily used for software interchange.
+
+  If distribution of object code is made by offering access to copy
+from a designated place, then offering equivalent access to copy the
+source code from the same place satisfies the requirement to
+distribute the source code, even though third parties are not
+compelled to copy the source along with the object code.
+
+  5. A program that contains no derivative of any portion of the
+Library, but is designed to work with the Library by being compiled or
+linked with it, is called a "work that uses the Library".  Such a
+work, in isolation, is not a derivative work of the Library, and
+therefore falls outside the scope of this License.
+
+  However, linking a "work that uses the Library" with the Library
+creates an executable that is a derivative of the Library (because it
+contains portions of the Library), rather than a "work that uses the
+library".  The executable is therefore covered by this License.
+Section 6 states terms for distribution of such executables.
+
+  When a "work that uses the Library" uses material from a header file
+that is part of the Library, the object code for the work may be a
+derivative work of the Library even though the source code is not.
+Whether this is true is especially significant if the work can be
+linked without the Library, or if the work is itself a library.  The
+threshold for this to be true is not precisely defined by law.
+
+  If such an object file uses only numerical parameters, data
+structure layouts and accessors, and small macros and small inline
+functions (ten lines or less in length), then the use of the object
+file is unrestricted, regardless of whether it is legally a derivative
+work.  (Executables containing this object code plus portions of the
+Library will still fall under Section 6.)
+
+  Otherwise, if the work is a derivative of the Library, you may
+distribute the object code for the work under the terms of Section 6.
+Any executables containing that work also fall under Section 6,
+whether or not they are linked directly with the Library itself.
+
+  6. As an exception to the Sections above, you may also compile or
+link a "work that uses the Library" with the Library to produce a
+work containing portions of the Library, and distribute that work
+under terms of your choice, provided that the terms permit
+modification of the work for the customer's own use and reverse
+engineering for debugging such modifications.
+
+  You must give prominent notice with each copy of the work that the
+Library is used in it and that the Library and its use are covered by
+this License.  You must supply a copy of this License.  If the work
+during execution displays copyright notices, you must include the
+copyright notice for the Library among them, as well as a reference
+directing the user to the copy of this License.  Also, you must do one
+of these things:
+
+    a) Accompany the work with the complete corresponding
+    machine-readable source code for the Library including whatever
+    changes were used in the work (which must be distributed under
+    Sections 1 and 2 above); and, if the work is an executable linked
+    with the Library, with the complete machine-readable "work that
+    uses the Library", as object code and/or source code, so that the
+    user can modify the Library and then relink to produce a modified
+    executable containing the modified Library.  (It is understood
+    that the user who changes the contents of definitions files in the
+    Library will not necessarily be able to recompile the application
+    to use the modified definitions.)
+
+    b) Accompany the work with a written offer, valid for at
+    least three years, to give the same user the materials
+    specified in Subsection 6a, above, for a charge no more
+    than the cost of performing this distribution.
+
+    c) If distribution of the work is made by offering access to copy
+    from a designated place, offer equivalent access to copy the above
+    specified materials from the same place.
+
+    d) Verify that the user has already received a copy of these
+    materials or that you have already sent this user a copy.
+
+  For an executable, the required form of the "work that uses the
+Library" must include any data and utility programs needed for
+reproducing the executable from it.  However, as a special exception,
+the source code distributed need not include anything that is normally
+distributed (in either source or binary form) with the major
+components (compiler, kernel, and so on) of the operating system on
+which the executable runs, unless that component itself accompanies
+the executable.
+
+  It may happen that this requirement contradicts the license
+restrictions of other proprietary libraries that do not normally
+accompany the operating system.  Such a contradiction means you cannot
+use both them and the Library together in an executable that you
+distribute.
+
+  7. You may place library facilities that are a work based on the
+Library side-by-side in a single library together with other library
+facilities not covered by this License, and distribute such a combined
+library, provided that the separate distribution of the work based on
+the Library and of the other library facilities is otherwise
+permitted, and provided that you do these two things:
+
+    a) Accompany the combined library with a copy of the same work
+    based on the Library, uncombined with any other library
+    facilities.  This must be distributed under the terms of the
+    Sections above.
+
+    b) Give prominent notice with the combined library of the fact
+    that part of it is a work based on the Library, and explaining
+    where to find the accompanying uncombined form of the same work.
+
+  8. You may not copy, modify, sublicense, link with, or distribute
+the Library except as expressly provided under this License.  Any
+attempt otherwise to copy, modify, sublicense, link with, or
+distribute the Library is void, and will automatically terminate your
+rights under this License.  However, parties who have received copies,
+or rights, from you under this License will not have their licenses
+terminated so long as such parties remain in full compliance.
+
+  9. You are not required to accept this License, since you have not
+signed it.  However, nothing else grants you permission to modify or
+distribute the Library or its derivative works.  These actions are
+prohibited by law if you do not accept this License.  Therefore, by
+modifying or distributing the Library (or any work based on the
+Library), you indicate your acceptance of this License to do so, and
+all its terms and conditions for copying, distributing or modifying
+the Library or works based on it.
+
+  10. Each time you redistribute the Library (or any work based on the
+Library), the recipient automatically receives a license from the
+original licensor to copy, distribute, link with or modify the Library
+subject to these terms and conditions.  You may not impose any further
+restrictions on the recipients' exercise of the rights granted herein.
+You are not responsible for enforcing compliance by third parties to
+this License.
+
+  11. If, as a consequence of a court judgment or allegation of patent
+infringement or for any other reason (not limited to patent issues),
+conditions are imposed on you (whether by court order, agreement or
+otherwise) that contradict the conditions of this License, they do not
+excuse you from the conditions of this License.  If you cannot
+distribute so as to satisfy simultaneously your obligations under this
+License and any other pertinent obligations, then as a consequence you
+may not distribute the Library at all.  For example, if a patent
+license would not permit royalty-free redistribution of the Library by
+all those who receive copies directly or indirectly through you, then
+the only way you could satisfy both it and this License would be to
+refrain entirely from distribution of the Library.
+
+If any portion of this section is held invalid or unenforceable under any
+particular circumstance, the balance of the section is intended to apply,
+and the section as a whole is intended to apply in other circumstances.
+
+It is not the purpose of this section to induce you to infringe any
+patents or other property right claims or to contest validity of any
+such claims; this section has the sole purpose of protecting the
+integrity of the free software distribution system which is
+implemented by public license practices.  Many people have made
+generous contributions to the wide range of software distributed
+through that system in reliance on consistent application of that
+system; it is up to the author/donor to decide if he or she is willing
+to distribute software through any other system and a licensee cannot
+impose that choice.
+
+This section is intended to make thoroughly clear what is believed to
+be a consequence of the rest of this License.
+
+  12. If the distribution and/or use of the Library is restricted in
+certain countries either by patents or by copyrighted interfaces, the
+original copyright holder who places the Library under this License may add
+an explicit geographical distribution limitation excluding those countries,
+so that distribution is permitted only in or among countries not thus
+excluded.  In such case, this License incorporates the limitation as if
+written in the body of this License.
+
+  13. The Free Software Foundation may publish revised and/or new
+versions of the Library General Public License from time to time.
+Such new versions will be similar in spirit to the present version,
+but may differ in detail to address new problems or concerns.
+
+Each version is given a distinguishing version number.  If the Library
+specifies a version number of this License which applies to it and
+"any later version", you have the option of following the terms and
+conditions either of that version or of any later version published by
+the Free Software Foundation.  If the Library does not specify a
+license version number, you may choose any version ever published by
+the Free Software Foundation.
+
+  14. If you wish to incorporate parts of the Library into other free
+programs whose distribution conditions are incompatible with these,
+write to the author to ask for permission.  For software which is
+copyrighted by the Free Software Foundation, write to the Free
+Software Foundation; we sometimes make exceptions for this.  Our
+decision will be guided by the two goals of preserving the free status
+of all derivatives of our free software and of promoting the sharing
+and reuse of software generally.
+
+			    NO WARRANTY
+
+  15. BECAUSE THE LIBRARY IS LICENSED FREE OF CHARGE, THERE IS NO
+WARRANTY FOR THE LIBRARY, TO THE EXTENT PERMITTED BY APPLICABLE LAW.
+EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT HOLDERS AND/OR
+OTHER PARTIES PROVIDE THE LIBRARY "AS IS" WITHOUT WARRANTY OF ANY
+KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+PURPOSE.  THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE
+LIBRARY IS WITH YOU.  SHOULD THE LIBRARY PROVE DEFECTIVE, YOU ASSUME
+THE COST OF ALL NECESSARY SERVICING, REPAIR OR CORRECTION.
+
+  16. IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN
+WRITING WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MAY MODIFY
+AND/OR REDISTRIBUTE THE LIBRARY AS PERMITTED ABOVE, BE LIABLE TO YOU
+FOR DAMAGES, INCLUDING ANY GENERAL, SPECIAL, INCIDENTAL OR
+CONSEQUENTIAL DAMAGES ARISING OUT OF THE USE OR INABILITY TO USE THE
+LIBRARY (INCLUDING BUT NOT LIMITED TO LOSS OF DATA OR DATA BEING
+RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD PARTIES OR A
+FAILURE OF THE LIBRARY TO OPERATE WITH ANY OTHER SOFTWARE), EVEN IF
+SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH
+DAMAGES.
+
+		     END OF TERMS AND CONDITIONS
+
+           How to Apply These Terms to Your New Libraries
+
+  If you develop a new library, and you want it to be of the greatest
+possible use to the public, we recommend making it free software that
+everyone can redistribute and change.  You can do so by permitting
+redistribution under these terms (or, alternatively, under the terms of the
+ordinary General Public License).
+
+  To apply these terms, attach the following notices to the library.  It is
+safest to attach them to the start of each source file to most effectively
+convey the exclusion of warranty; and each file should have at least the
+"copyright" line and a pointer to where the full notice is found.
+
+    <one line to give the library's name and a brief idea of what it does.>
+    Copyright (C) <year>  <name of author>
+
+    This library is free software; you can redistribute it and/or
+    modify it under the terms of the GNU Library General Public
+    License as published by the Free Software Foundation; either
+    version 2 of the License, or (at your option) any later version.
+
+    This library is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+    Library General Public License for more details.
+
+    You should have received a copy of the GNU Library General Public
+    License along with this library; if not, write to the Free Software
+    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+
+Also add information on how to contact you by electronic and paper mail.
+
+You should also get your employer (if you work as a programmer) or your
+school, if any, to sign a "copyright disclaimer" for the library, if
+necessary.  Here is a sample; alter the names:
+
+  Yoyodyne, Inc., hereby disclaims all copyright interest in the
+  library `Frob' (a library for tweaking knobs) written by James Random Hacker.
+
+  <signature of Ty Coon>, 1 April 1990
+  Ty Coon, President of Vice
+
+That's all there is to it!
diff --git a/surfsense_local/backend/bundling/licenses/lameenc/LAME-README b/surfsense_local/backend/bundling/licenses/lameenc/LAME-README
new file mode 100644
index 000000000..77c6c47ae
--- /dev/null
+++ b/surfsense_local/backend/bundling/licenses/lameenc/LAME-README
@@ -0,0 +1,44 @@
+                      LAME 3.xx
+               LAME Ain't an MP3 Encoder
+                  http://lame.sf.net
+	                  May 2011
+
+Originally developed by Mike Cheng (www.uq.net.au/~zzmcheng) and was
+latter developed by Mark Taylor (www.mp3dev.org). Currently maintained
+by The LAME Project.
+
+This code is distributed under the GNU LIBRARY GENERAL PUBLIC LICENSE
+(LGPL, see www.gnu.org), version 2.
+
+As LAME may contain software for which some companies may claim software
+patents, if you are in a location where software patents are recognized, it is
+suggested that you seek legal advice before deploying and/or redistributing
+LAME.
+
+In particular, it is suggested to visit
+
+    http://www.mp3licensing.com/
+
+if it applies to your jurisdiction.
+
+============================================================================
+
+see the file "INSTALL" for installation (compiling) instructions.
+see the file "USAGE" for the most up-to-date guide to the command line options.
+see the file "LICENSE" for details on how to use LAME in non-GPL programs.
+see the file "HACKING" if you are interested in working on LAME
+see the file "API" for details of the LAME encoding library API
+
+There is HTML documentation and a man page in the doc directory.
+
+============================================================================
+
+LAME uses the MPGLIB decoding engine, from the mpg123 package, written
+by: Michael Hipp (www.mpg123.de) MPGLIB is released under the GPL.
+
+Copyrights (c) 1999-2011 by The LAME Project
+Copyrights (c) 1999,2000,2001 by Mark Taylor
+Copyrights (c) 1998 by Michael Cheng
+Copyrights (c) 1995,1996,1997 by Michael Hipp: mpglib
+
+As well as additional copyrights as documented in the source code.
diff --git a/surfsense_local/backend/bundling/licenses/lameenc/SOURCE.txt b/surfsense_local/backend/bundling/licenses/lameenc/SOURCE.txt
new file mode 100644
index 000000000..f2591e951
--- /dev/null
+++ b/surfsense_local/backend/bundling/licenses/lameenc/SOURCE.txt
@@ -0,0 +1,23 @@
+lameenc and LAME in this application
+====================================
+
+The worker encodes podcast episodes to MP3 with lameenc, a Python extension
+module that statically links the LAME MP3 encoder. The module ships as its own
+file beside this folder, and may be replaced with a build of your own.
+
+lameenc 1.8.4, LGPL-3.0
+  Source: https://github.com/chrisstaite/lameenc/tree/v1.8.4
+  Licence: lameenc-LGPL-3.0.txt, with GPL-3.0.txt, which LGPL-3.0 builds on.
+
+LAME 3.100, LGPL (GNU Library General Public License, version 2)
+  Source: the official release, lame-3.100.tar.gz,
+  https://sourceforge.net/projects/lame/files/lame/3.100/
+  Licence: LAME-COPYING. Copyright notice: LAME-README.
+
+Where each text was taken from, unchanged:
+  lameenc-LGPL-3.0.txt  https://www.gnu.org/licenses/lgpl-3.0.txt
+  GPL-3.0.txt           https://www.gnu.org/licenses/gpl-3.0.txt
+  LAME-COPYING          COPYING in lame-3.100.tar.gz
+  LAME-README           README in lame-3.100.tar.gz
+  lame-3.100.tar.gz SHA-256:
+  ddfe36cab873794038ae2c1210557ad34857a4b6bdc515785d1da9e175b1da1e
diff --git a/surfsense_local/backend/bundling/licenses/lameenc/lameenc-LGPL-3.0.txt b/surfsense_local/backend/bundling/licenses/lameenc/lameenc-LGPL-3.0.txt
new file mode 100644
index 000000000..0a041280b
--- /dev/null
+++ b/surfsense_local/backend/bundling/licenses/lameenc/lameenc-LGPL-3.0.txt
@@ -0,0 +1,165 @@
+                   GNU LESSER GENERAL PUBLIC LICENSE
+                       Version 3, 29 June 2007
+
+ Copyright (C) 2007 Free Software Foundation, Inc. <https://fsf.org/>
+ Everyone is permitted to copy and distribute verbatim copies
+ of this license document, but changing it is not allowed.
+
+
+  This version of the GNU Lesser General Public License incorporates
+the terms and conditions of version 3 of the GNU General Public
+License, supplemented by the additional permissions listed below.
+
+  0. Additional Definitions.
+
+  As used herein, "this License" refers to version 3 of the GNU Lesser
+General Public License, and the "GNU GPL" refers to version 3 of the GNU
+General Public License.
+
+  "The Library" refers to a covered work governed by this License,
+other than an Application or a Combined Work as defined below.
+
+  An "Application" is any work that makes use of an interface provided
+by the Library, but which is not otherwise based on the Library.
+Defining a subclass of a class defined by the Library is deemed a mode
+of using an interface provided by the Library.
+
+  A "Combined Work" is a work produced by combining or linking an
+Application with the Library.  The particular version of the Library
+with which the Combined Work was made is also called the "Linked
+Version".
+
+  The "Minimal Corresponding Source" for a Combined Work means the
+Corresponding Source for the Combined Work, excluding any source code
+for portions of the Combined Work that, considered in isolation, are
+based on the Application, and not on the Linked Version.
+
+  The "Corresponding Application Code" for a Combined Work means the
+object code and/or source code for the Application, including any data
+and utility programs needed for reproducing the Combined Work from the
+Application, but excluding the System Libraries of the Combined Work.
+
+  1. Exception to Section 3 of the GNU GPL.
+
+  You may convey a covered work under sections 3 and 4 of this License
+without being bound by section 3 of the GNU GPL.
+
+  2. Conveying Modified Versions.
+
+  If you modify a copy of the Library, and, in your modifications, a
+facility refers to a function or data to be supplied by an Application
+that uses the facility (other than as an argument passed when the
+facility is invoked), then you may convey a copy of the modified
+version:
+
+   a) under this License, provided that you make a good faith effort to
+   ensure that, in the event an Application does not supply the
+   function or data, the facility still operates, and performs
+   whatever part of its purpose remains meaningful, or
+
+   b) under the GNU GPL, with none of the additional permissions of
+   this License applicable to that copy.
+
+  3. Object Code Incorporating Material from Library Header Files.
+
+  The object code form of an Application may incorporate material from
+a header file that is part of the Library.  You may convey such object
+code under terms of your choice, provided that, if the incorporated
+material is not limited to numerical parameters, data structure
+layouts and accessors, or small macros, inline functions and templates
+(ten or fewer lines in length), you do both of the following:
+
+   a) Give prominent notice with each copy of the object code that the
+   Library is used in it and that the Library and its use are
+   covered by this License.
+
+   b) Accompany the object code with a copy of the GNU GPL and this license
+   document.
+
+  4. Combined Works.
+
+  You may convey a Combined Work under terms of your choice that,
+taken together, effectively do not restrict modification of the
+portions of the Library contained in the Combined Work and reverse
+engineering for debugging such modifications, if you also do each of
+the following:
+
+   a) Give prominent notice with each copy of the Combined Work that
+   the Library is used in it and that the Library and its use are
+   covered by this License.
+
+   b) Accompany the Combined Work with a copy of the GNU GPL and this license
+   document.
+
+   c) For a Combined Work that displays copyright notices during
+   execution, include the copyright notice for the Library among
+   these notices, as well as a reference directing the user to the
+   copies of the GNU GPL and this license document.
+
+   d) Do one of the following:
+
+       0) Convey the Minimal Corresponding Source under the terms of this
+       License, and the Corresponding Application Code in a form
+       suitable for, and under terms that permit, the user to
+       recombine or relink the Application with a modified version of
+       the Linked Version to produce a modified Combined Work, in the
+       manner specified by section 6 of the GNU GPL for conveying
+       Corresponding Source.
+
+       1) Use a suitable shared library mechanism for linking with the
+       Library.  A suitable mechanism is one that (a) uses at run time
+       a copy of the Library already present on the user's computer
+       system, and (b) will operate properly with a modified version
+       of the Library that is interface-compatible with the Linked
+       Version.
+
+   e) Provide Installation Information, but only if you would otherwise
+   be required to provide such information under section 6 of the
+   GNU GPL, and only to the extent that such information is
+   necessary to install and execute a modified version of the
+   Combined Work produced by recombining or relinking the
+   Application with a modified version of the Linked Version. (If
+   you use option 4d0, the Installation Information must accompany
+   the Minimal Corresponding Source and Corresponding Application
+   Code. If you use option 4d1, you must provide the Installation
+   Information in the manner specified by section 6 of the GNU GPL
+   for conveying Corresponding Source.)
+
+  5. Combined Libraries.
+
+  You may place library facilities that are a work based on the
+Library side by side in a single library together with other library
+facilities that are not Applications and are not covered by this
+License, and convey such a combined library under terms of your
+choice, if you do both of the following:
+
+   a) Accompany the combined library with a copy of the same work based
+   on the Library, uncombined with any other library facilities,
+   conveyed under the terms of this License.
+
+   b) Give prominent notice with the combined library that part of it
+   is a work based on the Library, and explaining where to find the
+   accompanying uncombined form of the same work.
+
+  6. Revised Versions of the GNU Lesser General Public License.
+
+  The Free Software Foundation may publish revised and/or new versions
+of the GNU Lesser General Public License from time to time. Such new
+versions will be similar in spirit to the present version, but may
+differ in detail to address new problems or concerns.
+
+  Each version is given a distinguishing version number. If the
+Library as you received it specifies that a certain numbered version
+of the GNU Lesser General Public License "or any later version"
+applies to it, you have the option of following the terms and
+conditions either of that published version or of any later version
+published by the Free Software Foundation. If the Library as you
+received it does not specify a version number of the GNU Lesser
+General Public License, you may choose any version of the GNU Lesser
+General Public License ever published by the Free Software Foundation.
+
+  If the Library as you received it specifies that a proxy can decide
+whether future versions of the GNU Lesser General Public License shall
+apply, that proxy's public statement of acceptance of any version is
+permanent authorization for you to choose that version for the
+Library.
diff --git a/surfsense_local/backend/bundling/worker.spec b/surfsense_local/backend/bundling/worker.spec
index 0f7123ccc..316be3a64 100644
--- a/surfsense_local/backend/bundling/worker.spec
+++ b/surfsense_local/backend/bundling/worker.spec
@@ -85,9 +85,14 @@ binaries += collect_dynamic_libs("tokenizers")
 datas += collect_data_files("worker.studio", includes=["**/*.md"])
 datas += collect_data_files("modules.chat", includes=["prompts/*.md"])
 
-# lameenc, which encodes a podcast as MP3, is found by its import; the wheel
-# carries LAME's LGPL text only in its dist-info, which ships only when named.
+# lameenc, which encodes a podcast as MP3, is found by its import. Its
+# dist-info holds lameenc's LGPL-3.0 and ships only when named; the GPL-3.0 that
+# LGPL-3 §4(b) also asks for, LAME's licence and notice, and where both sources
+# are, ship from bundling/licenses/lameenc beside the module.
 datas += copy_metadata("lameenc")
+datas.append(
+    (str(BACKEND / "bundling" / "licenses" / "lameenc"), "lameenc-licenses")
+)
 
 # Huey resolves a task by its name, so the module that registers it must be in.
 hiddenimports += [
diff --git a/surfsense_local/backend/modules/llm/providers/audiocpp/speech.py b/surfsense_local/backend/modules/llm/providers/audiocpp/speech.py
index 0ba601200..84b21ba2a 100644
--- a/surfsense_local/backend/modules/llm/providers/audiocpp/speech.py
+++ b/surfsense_local/backend/modules/llm/providers/audiocpp/speech.py
@@ -11,13 +11,13 @@ import httpx
 
 from modules.llm.catalog.local.engines.audiocpp.manifest_fields import AudioDefaults
 from modules.llm.hardware import system_memory
-from modules.llm.providers.audiocpp.encoded_episode import encoded_episode
-from modules.llm.providers.audiocpp.joined_wav import joined_wav
 from modules.llm.providers.audiocpp.memory import (
     NotEnoughMemoryError,
     OtherModel,
     check_voicing_memory,
 )
+from modules.llm.providers.episode_audio.encoded_episode import encoded_episode
+from modules.llm.providers.episode_audio.joined_wav import joined_wav
 from modules.llm.providers.protocols import SpokenTurn, SynthesizedAudio, Voice
 from shared import cancellation
 
diff --git a/surfsense_local/backend/modules/llm/providers/episode_audio/__init__.py b/surfsense_local/backend/modules/llm/providers/episode_audio/__init__.py
new file mode 100644
index 000000000..48c24a18a
--- /dev/null
+++ b/surfsense_local/backend/modules/llm/providers/episode_audio/__init__.py
@@ -0,0 +1 @@
+"""A podcast episode's audio, whichever voice made its turns: joined, then encoded."""
diff --git a/surfsense_local/backend/modules/llm/providers/audiocpp/encoded_episode.py b/surfsense_local/backend/modules/llm/providers/episode_audio/encoded_episode.py
similarity index 88%
rename from surfsense_local/backend/modules/llm/providers/audiocpp/encoded_episode.py
rename to surfsense_local/backend/modules/llm/providers/episode_audio/encoded_episode.py
index fbeccc055..1fd882052 100644
--- a/surfsense_local/backend/modules/llm/providers/audiocpp/encoded_episode.py
+++ b/surfsense_local/backend/modules/llm/providers/episode_audio/encoded_episode.py
@@ -18,11 +18,11 @@ QUALITY = 2
 
 
 def encoded_episode(wav: bytes) -> SynthesizedAudio:
-    """`wav` is joined_wav's output: one format, 16-bit PCM from audio.cpp."""
+    """`wav` is joined_wav's output; LAME takes 16-bit PCM only."""
     try:
         import lameenc
     except ImportError:
-        logger.warning("audiocpp: lameenc is not installed; the episode stays WAV")
+        logger.warning("lameenc is not installed; the episode stays WAV")
         return SynthesizedAudio(wav, "audio/wav")
     with wave.open(io.BytesIO(wav)) as pcm:
         channels, width, rate = (
diff --git a/surfsense_local/backend/modules/llm/providers/audiocpp/joined_wav.py b/surfsense_local/backend/modules/llm/providers/episode_audio/joined_wav.py
similarity index 100%
rename from surfsense_local/backend/modules/llm/providers/audiocpp/joined_wav.py
rename to surfsense_local/backend/modules/llm/providers/episode_audio/joined_wav.py
diff --git a/surfsense_local/backend/modules/llm/providers/openai_compatible/joined_replies.py b/surfsense_local/backend/modules/llm/providers/openai_compatible/joined_replies.py
index 5e940bf23..fbde05d23 100644
--- a/surfsense_local/backend/modules/llm/providers/openai_compatible/joined_replies.py
+++ b/surfsense_local/backend/modules/llm/providers/openai_compatible/joined_replies.py
@@ -9,7 +9,7 @@ need the episode's encoder settings, which no reply states.
 import struct
 from collections.abc import Sequence
 
-from modules.llm.providers.audiocpp.joined_wav import joined_wav
+from modules.llm.providers.episode_audio.joined_wav import joined_wav
 from modules.llm.providers.protocols import SynthesizedAudio
 
 __all__ = ["joined_replies"]
diff --git a/surfsense_local/backend/modules/llm/providers/openai_compatible/speech.py b/surfsense_local/backend/modules/llm/providers/openai_compatible/speech.py
index 1f99d33cd..2d5760c91 100644
--- a/surfsense_local/backend/modules/llm/providers/openai_compatible/speech.py
+++ b/surfsense_local/backend/modules/llm/providers/openai_compatible/speech.py
@@ -1,11 +1,13 @@
 """Podcast voices from a server's /audio/speech, one request per turn."""
 
+import asyncio
 import logging
 import wave
 
 import httpx
 
 from modules.llm.connections.key_headers import key_headers
+from modules.llm.providers.episode_audio.encoded_episode import encoded_episode
 from modules.llm.providers.openai_compatible.joined_replies import joined_replies
 from modules.llm.providers.protocols import SpokenTurn, SynthesizedAudio
 from shared import cancellation
@@ -65,11 +67,15 @@ class RemoteSpeech:
                     )
                 voiced.append(reply.content)
         try:
-            return joined_replies(voiced)
+            joined = joined_replies(voiced)
         except (wave.Error, EOFError, ValueError) as error:
             raise NonRetryableSpeechError(
                 f"the server answered, but not as WAV or MP3: {error}"
             ) from error
+        if joined.media_type != "audio/wav":
+            return joined
+        # Off the event loop: encoding a long episode takes seconds of CPU.
+        return await asyncio.to_thread(encoded_episode, joined.content)
 
     async def _voice(
         self,
diff --git a/surfsense_local/backend/tests/unit/bundling/test_lameenc_licences.py b/surfsense_local/backend/tests/unit/bundling/test_lameenc_licences.py
new file mode 100644
index 000000000..759a003db
--- /dev/null
+++ b/surfsense_local/backend/tests/unit/bundling/test_lameenc_licences.py
@@ -0,0 +1,32 @@
+"""The worker ships what LGPL-3 §4 asks of lameenc and LAME beside the module."""
+
+from pathlib import Path
+
+import pytest
+
+pytestmark = pytest.mark.unit
+
+BUNDLING = Path(__file__).resolve().parents[3] / "bundling"
+LICENCES = BUNDLING / "licenses" / "lameenc"
+
+
+@pytest.mark.parametrize(
+    "name",
+    [
+        "lameenc-LGPL-3.0.txt",
+        "GPL-3.0.txt",
+        "LAME-COPYING",
+        "LAME-README",
+        "SOURCE.txt",
+    ],
+)
+def test_each_licence_text_is_there_and_not_empty(name: str) -> None:
+    """An empty file would ship a notice that satisfies nothing."""
+    assert (LICENCES / name).stat().st_size > 0
+
+
+def test_the_worker_bundles_the_licence_folder() -> None:
+    """The dist-info carries lameenc's LGPL only: no GPL, nothing of LAME's."""
+    spec = (BUNDLING / "worker.spec").read_text()
+
+    assert '"bundling" / "licenses" / "lameenc"' in spec
diff --git a/surfsense_local/backend/tests/unit/llm/providers/openai_compatible/test_speech.py b/surfsense_local/backend/tests/unit/llm/providers/openai_compatible/test_speech.py
index 6a5541677..9dab97fb5 100644
--- a/surfsense_local/backend/tests/unit/llm/providers/openai_compatible/test_speech.py
+++ b/surfsense_local/backend/tests/unit/llm/providers/openai_compatible/test_speech.py
@@ -3,6 +3,7 @@
 import asyncio
 import io
 import json
+import sys
 import wave
 
 import httpx
@@ -56,8 +57,10 @@ def speak(server: StubServer, turns: list[SpokenTurn]):
     return asyncio.run(speech.synthesize(turns, "en"))
 
 
-def test_each_turn_is_one_speech_request_joined_into_one_wav() -> None:
-    """WAV is asked for every time, because only WAV joins without an encoder."""
+def test_each_turn_is_one_speech_request_joined_into_one_wav(monkeypatch) -> None:
+    """WAV is asked for every time, because only WAV joins without an encoder.
+    Read from the WAV, which an episode stays when no encoder is installed."""
+    monkeypatch.setitem(sys.modules, "lameenc", None)
     server = StubServer()
 
     audio = speak(server, [SpokenTurn("coral", "Hi."), SpokenTurn("onyx", "Hello.")])
@@ -114,8 +117,11 @@ def test_a_server_that_answers_in_another_format_says_so() -> None:
         asyncio.run(speech.synthesize([SpokenTurn("coral", "Hi.")], "en"))
 
 
-def test_a_reply_of_several_wavs_back_to_back_joins_as_all_of_them() -> None:
+def test_a_reply_of_several_wavs_back_to_back_joins_as_all_of_them(
+    monkeypatch,
+) -> None:
     """Some servers stream a long line as one WAV per sentence in one body."""
+    monkeypatch.setitem(sys.modules, "lameenc", None)
 
     def stacked(request: httpx.Request) -> httpx.Response:
         return httpx.Response(200, content=wav(600) + wav(400))
@@ -132,6 +138,16 @@ def test_a_reply_of_several_wavs_back_to_back_joins_as_all_of_them() -> None:
         assert joined.getnframes() == 1000
 
 
+def test_a_wav_episode_is_encoded_as_mp3() -> None:
+    """A podcast is MP3 whichever voice made it, as the local one's is."""
+    audio = speak(
+        StubServer(), [SpokenTurn("coral", "Hi."), SpokenTurn("onyx", "Hello.")]
+    )
+
+    assert audio.media_type == "audio/mpeg"
+    assert audio.content[0] == 0xFF and audio.content[1] & 0xE0 == 0xE0
+
+
 # MPEG-1 Layer III, 128 kbit/s, 44.1 kHz, no padding: 417 bytes a frame.
 FRAME = b"\xff\xfb\x90\x00" + b"\x00" * 413
 
@@ -178,6 +194,7 @@ def test_a_server_that_refuses_wav_voices_the_episode_as_one_mp3() -> None:
 
     assert server.formats == ["wav", "mp3", "mp3"]
     assert audio.media_type == "audio/mpeg"
+    # The server's own frames, not encoded a second time.
     assert audio.content == FRAME * 6
 
 

From 8362bd18338441ceface3b2c5be33f284c2b9dba Mon Sep 17 00:00:00 2001
From: Cedric Karungu <ckarungu921@gmail.com>
Date: Wed, 14 Oct 2026 00:32:25 +0200
Subject: [PATCH 4/5] fix(local): keep an episode as WAV when its MP3 encode
 fails

---
 docs/architecture/studio.md                   |  4 ++--
 .../episode_audio/encoded_episode.py          | 23 +++++++++++--------
 .../openai_compatible/test_speech.py          | 22 ++++++++++++++++++
 3 files changed, 38 insertions(+), 11 deletions(-)

diff --git a/docs/architecture/studio.md b/docs/architecture/studio.md
index 4ef302ac9..df936ec27 100644
--- a/docs/architecture/studio.md
+++ b/docs/architecture/studio.md
@@ -34,7 +34,7 @@ The catalog is a tuple of twelve `Format` rows in [`formats.py`](../../surfsense
 | `mindmap` | Mind map | text_gen | `content/mindmap/` | none; the outline is the body |
 | `flashcards` | Flashcards | text_gen | `content/flashcards/` | deck JSON |
 | `quiz` | Quiz | text_gen | `content/quiz/` | quiz JSON |
-| `podcast` | Podcast | text_gen, audio_gen | `media/audio/podcast/` | MP3 audio |
+| `podcast` | Podcast | text_gen, audio_gen | `media/audio/podcast/` | MP3 audio, or WAV without a working encoder |
 | `image` | Image | image_gen, text_gen | `media/visual/image/` | the image |
 | `infographic` | Infographic | image_gen, text_gen | `media/visual/infographic/` | the image |
 
@@ -179,7 +179,7 @@ It keeps no `spec` and no `version`, so it lists as a document of its own and no
 - **The voices are the model's roster**, from its manifest entry: each voice with the languages it speaks, one for a Kokoro or Kitten voice and all 31 for a Supertonic voice. The brief's language list, A to Z by the name shown in the interface language, and each speaker's voice picker come from it, the picker grouped into female and male voices. The brief opens in American English where a voice speaks it, else in any English, else in the roster's first language, and a remembered brief the chosen model cannot voice falls back to that.
 - **Memory is checked before anything loads, twice.** The worker compares the operating system's available memory (`MemAvailable` on Linux, free plus inactive pages on macOS, `GlobalMemoryStatusEx` on Windows) with the model's peak measured while voicing with every chunk of text full, from its entry, plus 1 GiB, and refuses short of it: "Voicing needs about 3.5 GB free; this computer has 1.8 GB. Supertonic 3 needs about 1.6 GB." The second sentence names the first other curated audio model, in the manifest's order, that would fit, and is left out when none would. It checks first, before the chat model drafts anything, so a machine that cannot voice the episode does not spend minutes writing it. Short there, it gives up the text runtime through the API ([the image job's hold](local-models/admission.md#the-image-jobs-hold)), which lets running replies finish and unloads the chat model, checks again, and lets it go for drafting, because voicing frees that memory anyway: on 16 GB, Qwen3 1.7B resident left 3.3 GB free and refused Kokoro's 3.5 before a word was drafted. At voicing it gives the runtime up again and holds it until the last turn is voiced, so a chat sent meanwhile waits instead of loading the chat model back into memory voicing needs. The router reloads the chat model on the next request after either. Each check waits up to 10 seconds for memory the app is giving back before it refuses, because a starting podcast ends sd-server's idle minutes but Electron stops it only on its next 5-second poll ([the image path](#the-image-path)). audio.cpp's own guard counts only the file it reads.
 - **One speech request per turn**, with the turn's voice. The brief's language goes in the request only for a voice that speaks several; a Kokoro voice's name already fixes its language. The turns are joined with 0.35 s between them.
-- **The episode is MP3**: the joined 16-bit WAV is encoded in-process by LAME through `lameenc`, at 64 kbit/s per channel and the model's own channels and rate, about a sixth of the WAV ([`encoded_episode.py`](../../surfsense_local/backend/modules/llm/providers/episode_audio/encoded_episode.py)), which a server's WAV episode goes through too. No ffmpeg and nothing on the user's `PATH` is used. Where `lameenc` cannot be imported, as in a checkout that skipped it, the episode stays WAV, which the viewer's `<audio>` plays the same.
+- **The episode is MP3**: the joined 16-bit WAV is encoded in-process by LAME through `lameenc`, at 64 kbit/s per channel and the model's own channels and rate, about a sixth of the WAV ([`encoded_episode.py`](../../surfsense_local/backend/modules/llm/providers/episode_audio/encoded_episode.py)), which a server's WAV episode goes through too. No ffmpeg and nothing on the user's `PATH` is used. Where `lameenc` cannot be imported, as in a checkout that skipped it, or its encode fails, the episode stays WAV, which the viewer's `<audio>` plays the same: the voices are paid for by then, so the episode is kept rather than failed and voiced again.
 - **A turn the server fails ends the episode with the server's own words**: "audio.cpp could not voice turn 3 of 40: unknown Kokoro voice id: nobody". The server was reached, so this is not "The model could not be reached", which stays for a server that does not answer.
 - **The model is unloaded when voicing ends**, with `POST /v1/tasks/unload_all_models`, success or failure, so its memory is back before the chat model's next job. The server's five-minute idle unload is the backstop.
 
diff --git a/surfsense_local/backend/modules/llm/providers/episode_audio/encoded_episode.py b/surfsense_local/backend/modules/llm/providers/episode_audio/encoded_episode.py
index 1fd882052..25a7babf5 100644
--- a/surfsense_local/backend/modules/llm/providers/episode_audio/encoded_episode.py
+++ b/surfsense_local/backend/modules/llm/providers/episode_audio/encoded_episode.py
@@ -1,5 +1,5 @@
 """The joined episode as MP3, about a sixth of its WAV, through LAME (lameenc).
-WAV where lameenc is not installed, so a checkout without it still plays."""
+WAV where lameenc is not installed or fails, so the episode is never lost."""
 
 import io
 import logging
@@ -34,11 +34,16 @@ def encoded_episode(wav: bytes) -> SynthesizedAudio:
     # LAME takes 16-bit samples only.
     if width != 2:
         return SynthesizedAudio(wav, "audio/wav")
-    encoder = lameenc.Encoder()
-    encoder.set_channels(channels)
-    encoder.set_in_sample_rate(rate)
-    encoder.set_bit_rate(KBPS_PER_CHANNEL * channels)
-    encoder.set_quality(QUALITY)
-    return SynthesizedAudio(
-        bytes(encoder.encode(frames) + encoder.flush()), "audio/mpeg"
-    )
+    try:
+        encoder = lameenc.Encoder()
+        encoder.set_channels(channels)
+        encoder.set_in_sample_rate(rate)
+        encoder.set_bit_rate(KBPS_PER_CHANNEL * channels)
+        encoder.set_quality(QUALITY)
+        mp3 = bytes(encoder.encode(frames) + encoder.flush())
+    except RuntimeError:
+        # The voices are paid for: keep the episode rather than fail a job
+        # whose retry would voice, and bill, it again.
+        logger.warning("MP3 encoding failed; the episode stays WAV", exc_info=True)
+        return SynthesizedAudio(wav, "audio/wav")
+    return SynthesizedAudio(mp3, "audio/mpeg")
diff --git a/surfsense_local/backend/tests/unit/llm/providers/openai_compatible/test_speech.py b/surfsense_local/backend/tests/unit/llm/providers/openai_compatible/test_speech.py
index 9dab97fb5..51ee40d3a 100644
--- a/surfsense_local/backend/tests/unit/llm/providers/openai_compatible/test_speech.py
+++ b/surfsense_local/backend/tests/unit/llm/providers/openai_compatible/test_speech.py
@@ -148,6 +148,28 @@ def test_a_wav_episode_is_encoded_as_mp3() -> None:
     assert audio.content[0] == 0xFF and audio.content[1] & 0xE0 == 0xE0
 
 
+def test_an_episode_the_encoder_fails_on_is_kept_as_wav(monkeypatch) -> None:
+    """The voices are paid for by then: the episode stays, as WAV, rather than a
+    failure the job would retry and bill twice."""
+    import lameenc
+
+    class Broken:
+        def __getattr__(self, name: str):
+            def fail(*args: object) -> None:
+                raise RuntimeError("lame_encode_buffer failed")
+
+            return fail
+
+    monkeypatch.setattr(lameenc, "Encoder", Broken)
+    audio = speak(
+        StubServer(), [SpokenTurn("coral", "Hi."), SpokenTurn("onyx", "Hello.")]
+    )
+
+    assert audio.media_type == "audio/wav"
+    with wave.open(io.BytesIO(audio.content)) as joined:
+        assert joined.getnframes() > 0
+
+
 # MPEG-1 Layer III, 128 kbit/s, 44.1 kHz, no padding: 417 bytes a frame.
 FRAME = b"\xff\xfb\x90\x00" + b"\x00" * 413
 

From 46cfe86190ee2e09a59adb7bb00d6739e4aab695 Mon Sep 17 00:00:00 2001
From: Cedric Karungu <ckarungu921@gmail.com>
Date: Wed, 14 Oct 2026 01:35:22 +0200
Subject: [PATCH 5/5] fix(local): encode the episode in the podcast pipeline,
 keeping each voice's WAV

---
 docs/architecture/studio.md                   |  2 +-
 .../modules/llm/providers/audiocpp/speech.py  |  4 +-
 .../episode_audio/encoded_episode.py          | 20 +++---
 .../llm/providers/openai_compatible/speech.py |  7 +-
 .../tests/packaging/test_audiocpp_voicing.py  | 24 ++-----
 .../llm/providers/audiocpp/test_speech.py     | 20 +-----
 .../llm/providers/episode_audio/__init__.py   |  0
 .../episode_audio/test_encoded_episode.py     | 71 +++++++++++++++++++
 .../openai_compatible/test_speech.py          | 28 +-------
 .../tests/unit/worker/test_studio_podcast.py  | 32 +++++++++
 .../studio/media/audio/podcast/pipeline.py    |  5 ++
 11 files changed, 133 insertions(+), 80 deletions(-)
 create mode 100644 surfsense_local/backend/tests/unit/llm/providers/episode_audio/__init__.py
 create mode 100644 surfsense_local/backend/tests/unit/llm/providers/episode_audio/test_encoded_episode.py

diff --git a/docs/architecture/studio.md b/docs/architecture/studio.md
index df936ec27..ef3e8f7ca 100644
--- a/docs/architecture/studio.md
+++ b/docs/architecture/studio.md
@@ -179,7 +179,7 @@ It keeps no `spec` and no `version`, so it lists as a document of its own and no
 - **The voices are the model's roster**, from its manifest entry: each voice with the languages it speaks, one for a Kokoro or Kitten voice and all 31 for a Supertonic voice. The brief's language list, A to Z by the name shown in the interface language, and each speaker's voice picker come from it, the picker grouped into female and male voices. The brief opens in American English where a voice speaks it, else in any English, else in the roster's first language, and a remembered brief the chosen model cannot voice falls back to that.
 - **Memory is checked before anything loads, twice.** The worker compares the operating system's available memory (`MemAvailable` on Linux, free plus inactive pages on macOS, `GlobalMemoryStatusEx` on Windows) with the model's peak measured while voicing with every chunk of text full, from its entry, plus 1 GiB, and refuses short of it: "Voicing needs about 3.5 GB free; this computer has 1.8 GB. Supertonic 3 needs about 1.6 GB." The second sentence names the first other curated audio model, in the manifest's order, that would fit, and is left out when none would. It checks first, before the chat model drafts anything, so a machine that cannot voice the episode does not spend minutes writing it. Short there, it gives up the text runtime through the API ([the image job's hold](local-models/admission.md#the-image-jobs-hold)), which lets running replies finish and unloads the chat model, checks again, and lets it go for drafting, because voicing frees that memory anyway: on 16 GB, Qwen3 1.7B resident left 3.3 GB free and refused Kokoro's 3.5 before a word was drafted. At voicing it gives the runtime up again and holds it until the last turn is voiced, so a chat sent meanwhile waits instead of loading the chat model back into memory voicing needs. The router reloads the chat model on the next request after either. Each check waits up to 10 seconds for memory the app is giving back before it refuses, because a starting podcast ends sd-server's idle minutes but Electron stops it only on its next 5-second poll ([the image path](#the-image-path)). audio.cpp's own guard counts only the file it reads.
 - **One speech request per turn**, with the turn's voice. The brief's language goes in the request only for a voice that speaks several; a Kokoro voice's name already fixes its language. The turns are joined with 0.35 s between them.
-- **The episode is MP3**: the joined 16-bit WAV is encoded in-process by LAME through `lameenc`, at 64 kbit/s per channel and the model's own channels and rate, about a sixth of the WAV ([`encoded_episode.py`](../../surfsense_local/backend/modules/llm/providers/episode_audio/encoded_episode.py)), which a server's WAV episode goes through too. No ffmpeg and nothing on the user's `PATH` is used. Where `lameenc` cannot be imported, as in a checkout that skipped it, or its encode fails, the episode stays WAV, which the viewer's `<audio>` plays the same: the voices are paid for by then, so the episode is kept rather than failed and voiced again.
+- **The episode is MP3**: the joined 16-bit WAV is encoded in-process by LAME through `lameenc`, at 64 kbit/s per channel and the model's own channels and rate, about a sixth of the WAV ([`encoded_episode.py`](../../surfsense_local/backend/modules/llm/providers/episode_audio/encoded_episode.py)). The podcast pipeline encodes whatever WAV a voice returns, local or a server's, so a voice's `synthesize` stays WAV for every other caller, such as Settings' voice check, which plays it as it came. No ffmpeg and nothing on the user's `PATH` is used. Where `lameenc` cannot be imported, as in a checkout that skipped it, or its encode fails, the episode stays WAV, which the viewer's `<audio>` plays the same: the voices are paid for by then, so the episode is kept rather than failed and voiced again.
 - **A turn the server fails ends the episode with the server's own words**: "audio.cpp could not voice turn 3 of 40: unknown Kokoro voice id: nobody". The server was reached, so this is not "The model could not be reached", which stays for a server that does not answer.
 - **The model is unloaded when voicing ends**, with `POST /v1/tasks/unload_all_models`, success or failure, so its memory is back before the chat model's next job. The server's five-minute idle unload is the backstop.
 
diff --git a/surfsense_local/backend/modules/llm/providers/audiocpp/speech.py b/surfsense_local/backend/modules/llm/providers/audiocpp/speech.py
index 84b21ba2a..e42f7ea9c 100644
--- a/surfsense_local/backend/modules/llm/providers/audiocpp/speech.py
+++ b/surfsense_local/backend/modules/llm/providers/audiocpp/speech.py
@@ -16,7 +16,6 @@ from modules.llm.providers.audiocpp.memory import (
     OtherModel,
     check_voicing_memory,
 )
-from modules.llm.providers.episode_audio.encoded_episode import encoded_episode
 from modules.llm.providers.episode_audio.joined_wav import joined_wav
 from modules.llm.providers.protocols import SpokenTurn, SynthesizedAudio, Voice
 from shared import cancellation
@@ -109,8 +108,7 @@ class AudioCppSpeech:
                     voiced = await self._voice_each(client, turns, language, speaks)
                 finally:
                     await _unload(client)
-        # Off the event loop: encoding a long episode takes seconds of CPU.
-        return await asyncio.to_thread(lambda: encoded_episode(joined_wav(voiced)))
+        return SynthesizedAudio(joined_wav(voiced), "audio/wav")
 
     async def _wait_for_memory(self, seconds: float) -> None:
         deadline = time.monotonic() + seconds
diff --git a/surfsense_local/backend/modules/llm/providers/episode_audio/encoded_episode.py b/surfsense_local/backend/modules/llm/providers/episode_audio/encoded_episode.py
index 25a7babf5..de5c9506d 100644
--- a/surfsense_local/backend/modules/llm/providers/episode_audio/encoded_episode.py
+++ b/surfsense_local/backend/modules/llm/providers/episode_audio/encoded_episode.py
@@ -24,13 +24,17 @@ def encoded_episode(wav: bytes) -> SynthesizedAudio:
     except ImportError:
         logger.warning("lameenc is not installed; the episode stays WAV")
         return SynthesizedAudio(wav, "audio/wav")
-    with wave.open(io.BytesIO(wav)) as pcm:
-        channels, width, rate = (
-            pcm.getnchannels(),
-            pcm.getsampwidth(),
-            pcm.getframerate(),
-        )
-        frames = pcm.readframes(pcm.getnframes())
+    try:
+        with wave.open(io.BytesIO(wav)) as pcm:
+            channels, width, rate = (
+                pcm.getnchannels(),
+                pcm.getsampwidth(),
+                pcm.getframerate(),
+            )
+            frames = pcm.readframes(pcm.getnframes())
+    except (wave.Error, EOFError):
+        logger.warning("the episode is not a WAV LAME can read; it stays as it is")
+        return SynthesizedAudio(wav, "audio/wav")
     # LAME takes 16-bit samples only.
     if width != 2:
         return SynthesizedAudio(wav, "audio/wav")
@@ -41,7 +45,7 @@ def encoded_episode(wav: bytes) -> SynthesizedAudio:
         encoder.set_bit_rate(KBPS_PER_CHANNEL * channels)
         encoder.set_quality(QUALITY)
         mp3 = bytes(encoder.encode(frames) + encoder.flush())
-    except RuntimeError:
+    except (RuntimeError, MemoryError):
         # The voices are paid for: keep the episode rather than fail a job
         # whose retry would voice, and bill, it again.
         logger.warning("MP3 encoding failed; the episode stays WAV", exc_info=True)
diff --git a/surfsense_local/backend/modules/llm/providers/openai_compatible/speech.py b/surfsense_local/backend/modules/llm/providers/openai_compatible/speech.py
index 2d5760c91..6284dac2a 100644
--- a/surfsense_local/backend/modules/llm/providers/openai_compatible/speech.py
+++ b/surfsense_local/backend/modules/llm/providers/openai_compatible/speech.py
@@ -1,13 +1,11 @@
 """Podcast voices from a server's /audio/speech, one request per turn."""
 
-import asyncio
 import logging
 import wave
 
 import httpx
 
 from modules.llm.connections.key_headers import key_headers
-from modules.llm.providers.episode_audio.encoded_episode import encoded_episode
 from modules.llm.providers.openai_compatible.joined_replies import joined_replies
 from modules.llm.providers.protocols import SpokenTurn, SynthesizedAudio
 from shared import cancellation
@@ -72,10 +70,7 @@ class RemoteSpeech:
             raise NonRetryableSpeechError(
                 f"the server answered, but not as WAV or MP3: {error}"
             ) from error
-        if joined.media_type != "audio/wav":
-            return joined
-        # Off the event loop: encoding a long episode takes seconds of CPU.
-        return await asyncio.to_thread(encoded_episode, joined.content)
+        return joined
 
     async def _voice(
         self,
diff --git a/surfsense_local/backend/tests/packaging/test_audiocpp_voicing.py b/surfsense_local/backend/tests/packaging/test_audiocpp_voicing.py
index 89a9c45c7..b0f66e056 100644
--- a/surfsense_local/backend/tests/packaging/test_audiocpp_voicing.py
+++ b/surfsense_local/backend/tests/packaging/test_audiocpp_voicing.py
@@ -11,12 +11,14 @@ fails the run: CI sets the folder, and a skip there would pass unseen.
 import asyncio
 import contextlib
 import hashlib
+import io
 import os
 import shutil
 import socket
 import subprocess
 import sys
 import time
+import wave
 from collections.abc import Iterator
 from pathlib import Path
 
@@ -201,24 +203,8 @@ def test_a_curated_model_voices_through_the_app_and_is_given_back(
         sys.stderr.write("\n".join(["audiocpp_server's log, last lines:", *tail]))
         raise
 
-    assert voiced.media_type == "audio/mpeg"
-    assert _mp3_rate(voiced.content) == audio.sample_rate
-    # At least a second of speech at 64 kbit/s per channel.
-    assert len(voiced.content) > 64_000 // 8
+    with wave.open(io.BytesIO(voiced.content)) as joined:
+        assert joined.getframerate() == audio.sample_rate
+        assert joined.getnframes() / joined.getframerate() > 1.0
     listed = httpx.get(f"{server}/v1/models").json()["data"]
     assert not [m["id"] for m in listed if m["loaded"]]
-
-
-# The sample rates an MPEG audio frame header can name, by version.
-_MPEG_RATES = {
-    3: (44100, 48000, 32000),
-    2: (22050, 24000, 16000),
-    0: (11025, 12000, 8000),
-}
-
-
-def _mp3_rate(episode: bytes) -> int:
-    """The sample rate the episode's first MPEG frame header names."""
-    assert episode[0] == 0xFF and episode[1] & 0xE0 == 0xE0, "an MPEG frame first"
-    version = (episode[1] >> 3) & 0b11
-    return _MPEG_RATES[version][(episode[2] >> 2) & 0b11]
diff --git a/surfsense_local/backend/tests/unit/llm/providers/audiocpp/test_speech.py b/surfsense_local/backend/tests/unit/llm/providers/audiocpp/test_speech.py
index ad39995f0..1b6529c59 100644
--- a/surfsense_local/backend/tests/unit/llm/providers/audiocpp/test_speech.py
+++ b/surfsense_local/backend/tests/unit/llm/providers/audiocpp/test_speech.py
@@ -170,28 +170,12 @@ def test_each_turn_is_one_request_and_the_turns_join_with_a_pause(monkeypatch) -
         assert joined.getnframes() == 10_400
 
 
-def test_the_episode_is_mp3_at_the_models_rate() -> None:
-    """About a sixth of the WAV, and the browser plays it as it did the WAV."""
+def test_the_voice_returns_its_joined_wav() -> None:
+    """MP3 is the podcast's own step; Settings' voice check plays this WAV."""
     turns = [SpokenTurn("am_adam", "Welcome back."), SpokenTurn("af_heart", "Thanks.")]
 
     audio = speak(voiced("kokoro-82m", "kokoro-82m-q8_0"), StubServer(24000), turns)
 
-    assert audio.media_type == "audio/mpeg"
-    header = audio.content[:4]
-    assert header[0] == 0xFF and header[1] & 0xE0 == 0xE0, "an MPEG frame first"
-    # MPEG-2's rate index 1 is 24 kHz, mono is channel mode 3.
-    assert (header[1] >> 3 & 0b11, header[2] >> 2 & 0b11) == (0b10, 1)
-    assert header[3] >> 6 == 0b11
-    assert len(audio.content) < len(wav(2 * 24000)) / 4
-
-
-def test_the_episode_stays_wav_without_an_encoder(monkeypatch) -> None:
-    """A checkout that skipped the dependency still gets an episode it can play."""
-    monkeypatch.setitem(sys.modules, "lameenc", None)
-    turns = [SpokenTurn("am_adam", "Welcome back."), SpokenTurn("af_heart", "Thanks.")]
-
-    audio = speak(voiced("kokoro-82m", "kokoro-82m-q8_0"), StubServer(), turns)
-
     assert audio.media_type == "audio/wav"
     assert audio.content[:4] == b"RIFF"
 
diff --git a/surfsense_local/backend/tests/unit/llm/providers/episode_audio/__init__.py b/surfsense_local/backend/tests/unit/llm/providers/episode_audio/__init__.py
new file mode 100644
index 000000000..e69de29bb
diff --git a/surfsense_local/backend/tests/unit/llm/providers/episode_audio/test_encoded_episode.py b/surfsense_local/backend/tests/unit/llm/providers/episode_audio/test_encoded_episode.py
new file mode 100644
index 000000000..ee19c24f7
--- /dev/null
+++ b/surfsense_local/backend/tests/unit/llm/providers/episode_audio/test_encoded_episode.py
@@ -0,0 +1,71 @@
+"""A podcast episode as MP3, or the WAV it came as whenever that cannot be done."""
+
+import io
+import sys
+import wave
+
+import pytest
+
+from modules.llm.providers.episode_audio.encoded_episode import encoded_episode
+
+pytestmark = pytest.mark.unit
+
+
+def wav(seconds: float = 2.0, rate: int = 24000) -> bytes:
+    """A voice's joined episode: 16-bit mono PCM."""
+    out = io.BytesIO()
+    with wave.open(out, "wb") as w:
+        w.setnchannels(1)
+        w.setsampwidth(2)
+        w.setframerate(rate)
+        w.writeframes(b"\x01\x00" * int(rate * seconds))
+    return out.getvalue()
+
+
+def test_an_episode_is_mp3_at_its_own_rate() -> None:
+    """About a sixth of the WAV, and the browser plays it as it did the WAV."""
+    audio = encoded_episode(wav())
+
+    assert audio.media_type == "audio/mpeg"
+    header = audio.content[:4]
+    assert header[0] == 0xFF and header[1] & 0xE0 == 0xE0, "an MPEG frame first"
+    # MPEG-2's rate index 1 is 24 kHz, mono is channel mode 3.
+    assert (header[1] >> 3 & 0b11, header[2] >> 2 & 0b11) == (0b10, 1)
+    assert header[3] >> 6 == 0b11
+    assert len(audio.content) < len(wav()) / 4
+
+
+def test_without_lameenc_the_episode_stays_wav(monkeypatch) -> None:
+    """A checkout that skipped the dependency still gets an episode it can play."""
+    monkeypatch.setitem(sys.modules, "lameenc", None)
+
+    audio = encoded_episode(wav())
+
+    assert (audio.media_type, audio.content[:4]) == ("audio/wav", b"RIFF")
+
+
+@pytest.mark.parametrize("raised", [RuntimeError, MemoryError])
+def test_an_episode_the_encoder_fails_on_stays_wav(monkeypatch, raised) -> None:
+    """The voices are paid for by then: the episode is kept, not failed and voiced again."""
+    import lameenc
+
+    class Broken:
+        def __getattr__(self, name: str):
+            def fail(*args: object) -> None:
+                raise raised("lame_encode_buffer failed")
+
+            return fail
+
+    monkeypatch.setattr(lameenc, "Encoder", Broken)
+    episode = wav()
+
+    audio = encoded_episode(episode)
+
+    assert (audio.media_type, audio.content) == ("audio/wav", episode)
+
+
+def test_bytes_that_are_not_a_wav_are_left_as_they_are() -> None:
+    """Nothing to encode from, so nothing is lost either."""
+    audio = encoded_episode(b"RIFFfake")
+
+    assert (audio.media_type, audio.content) == ("audio/wav", b"RIFFfake")
diff --git a/surfsense_local/backend/tests/unit/llm/providers/openai_compatible/test_speech.py b/surfsense_local/backend/tests/unit/llm/providers/openai_compatible/test_speech.py
index 51ee40d3a..3ba5cf846 100644
--- a/surfsense_local/backend/tests/unit/llm/providers/openai_compatible/test_speech.py
+++ b/surfsense_local/backend/tests/unit/llm/providers/openai_compatible/test_speech.py
@@ -138,36 +138,14 @@ def test_a_reply_of_several_wavs_back_to_back_joins_as_all_of_them(
         assert joined.getnframes() == 1000
 
 
-def test_a_wav_episode_is_encoded_as_mp3() -> None:
-    """A podcast is MP3 whichever voice made it, as the local one's is."""
-    audio = speak(
-        StubServer(), [SpokenTurn("coral", "Hi."), SpokenTurn("onyx", "Hello.")]
-    )
-
-    assert audio.media_type == "audio/mpeg"
-    assert audio.content[0] == 0xFF and audio.content[1] & 0xE0 == 0xE0
-
-
-def test_an_episode_the_encoder_fails_on_is_kept_as_wav(monkeypatch) -> None:
-    """The voices are paid for by then: the episode stays, as WAV, rather than a
-    failure the job would retry and bill twice."""
-    import lameenc
-
-    class Broken:
-        def __getattr__(self, name: str):
-            def fail(*args: object) -> None:
-                raise RuntimeError("lame_encode_buffer failed")
-
-            return fail
-
-    monkeypatch.setattr(lameenc, "Encoder", Broken)
+def test_a_wav_episode_is_returned_as_wav() -> None:
+    """MP3 is the podcast's own step; Settings' voice check plays this WAV."""
     audio = speak(
         StubServer(), [SpokenTurn("coral", "Hi."), SpokenTurn("onyx", "Hello.")]
     )
 
     assert audio.media_type == "audio/wav"
-    with wave.open(io.BytesIO(audio.content)) as joined:
-        assert joined.getnframes() > 0
+    assert audio.content[:4] == b"RIFF"
 
 
 # MPEG-1 Layer III, 128 kbit/s, 44.1 kHz, no padding: 417 bytes a frame.
diff --git a/surfsense_local/backend/tests/unit/worker/test_studio_podcast.py b/surfsense_local/backend/tests/unit/worker/test_studio_podcast.py
index 77f62c65f..46dfddc76 100644
--- a/surfsense_local/backend/tests/unit/worker/test_studio_podcast.py
+++ b/surfsense_local/backend/tests/unit/worker/test_studio_podcast.py
@@ -1,5 +1,7 @@
 """The podcast pipeline: a reviewed brief becomes an outline, then segments."""
 
+import io
+import wave
 from collections.abc import AsyncIterator
 from types import SimpleNamespace
 
@@ -303,6 +305,36 @@ def test_an_episode_is_planned_then_drafted_per_segment_then_voiced_per_speaker(
     assert "**Lee:** Goodbye." in built.markdown
 
 
+def _real_wav(seconds: float = 1.0, rate: int = 24000) -> bytes:
+    """A voice's joined episode: 16-bit mono PCM."""
+    out = io.BytesIO()
+    with wave.open(out, "wb") as w:
+        w.setnchannels(1)
+        w.setsampwidth(2)
+        w.setframerate(rate)
+        w.writeframes(b"\x01\x00" * int(rate * seconds))
+    return out.getvalue()
+
+
+def test_a_wav_episode_is_stored_as_mp3(monkeypatch: pytest.MonkeyPatch) -> None:
+    """The podcast is encoded here, whichever voice made it: synthesize stays
+    WAV for every other caller, such as Settings' voice check."""
+    voice, _ = _episode(
+        monkeypatch,
+        '{"title": "Saturn Rings", "segments": [{"title": "Open"}]}',
+        '{"turns": [{"speaker": 1, "text": "Hi."}, {"speaker": 2, "text": "Bye."}]}',
+        voice=FakeVoice(_real_wav(), "audio/wav"),
+    )
+
+    built = pipeline.render(MODEL, voice, [], None, BRIEF.model_dump(mode="json"))
+
+    assert (built.primary_mime, built.primary_filename) == (
+        "audio/mpeg",
+        "saturn-rings.mp3",
+    )
+    assert built.primary[0] == 0xFF and built.primary[1] & 0xE0 == 0xE0
+
+
 def test_an_mp3_episode_is_named_mp3(monkeypatch: pytest.MonkeyPatch) -> None:
     """The local voice encodes MP3; the stored file says so by its name."""
     voice, _ = _episode(
diff --git a/surfsense_local/backend/worker/studio/media/audio/podcast/pipeline.py b/surfsense_local/backend/worker/studio/media/audio/podcast/pipeline.py
index 49343d807..f5cd39ad4 100644
--- a/surfsense_local/backend/worker/studio/media/audio/podcast/pipeline.py
+++ b/surfsense_local/backend/worker/studio/media/audio/podcast/pipeline.py
@@ -2,6 +2,7 @@ import asyncio
 import logging
 
 from modules.artifacts.podcast.brief import PodcastBrief, Speaker
+from modules.llm.providers.episode_audio.encoded_episode import encoded_episode
 from modules.llm.providers.protocols import SpokenTurn, TextToSpeech
 from modules.llm.resolution import ResolvedGeneration
 from worker.studio.media.audio.podcast import draft, outline
@@ -47,6 +48,10 @@ def render(
         sum(len(turn.text) for turn in spoken),
     )
     audio = asyncio.run(voice.synthesize(spoken, brief.language))
+    # Encoded here, not by the voice: only a podcast is worth making MP3, and
+    # Settings' voice check plays the voice's own WAV.
+    if audio.media_type == "audio/wav":
+        audio = encoded_episode(audio.content)
     return Built(
         title=plan.title,
         markdown=_transcript(plan.title, brief, turns),