AudioSumma/summarize.py

119 lines
No EOL
4.3 KiB
Python

import os
import requests
import datetime
import tempfile
import subprocess
from dotenv import load_dotenv
# Use local models with the OpenAI library and a custom baseurl
from openai import OpenAI
# Load environment variables from .env file
load_dotenv()
# Load settings from environment
WHISPERCPP_URL = os.getenv("WHISPERCPP_URL")
LLAMACPP_URL = os.getenv("LLAMACPP_URL")
SYSTEM_MESSAGE = os.getenv("SYSTEM_MESSAGE")
SUMMARY_PROMPT = os.getenv("SUMMARY_PROMPT")
FACT_PROMPT = os.getenv("FACT_PROMPT")
SENTIMENT_PROMPT = os.getenv("SENTIMENT_PROMPT")
CHUNK_SIZE = int(os.getenv("CHUNK_SIZE"))
TEMPERATURE = float(os.getenv("TEMPERATURE"))
TOP_P = float(os.getenv("TOP_P"))
MAX_TOKENS = float(os.getenv("MAX_TOKENS"))
def whisper_api(file):
"""Transcribe audio file using Whisper API."""
files = {"file": file}
api_data = {
"temperature": "0.0",
"response_format": "json"
}
response = requests.post(WHISPERCPP_URL, data=api_data, files=files)
return response.json()["text"]
def llm_local(prompt):
client = OpenAI(api_key="doesntmatter", base_url=LLAMACPP_URL)
messages=[{"role": "system", "content": SYSTEM_MESSAGE},{"role": "user", "content": prompt}]
response = client.chat.completions.create(model="whatever", max_tokens=MAX_TOKENS, temperature=TEMPERATURE, top_p=TOP_P, messages=messages)
return response.choices[0].message.content
def trim_silence(filename):
"""Trim silence from audio file using FFmpeg."""
with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as temp_file:
temp_filename = temp_file.name
ffmpeg_command = [
"ffmpeg",
"-i", filename,
"-af", "silenceremove=stop_threshold=-40dB:stop_duration=1:stop_periods=-1",
"-y", # Overwrite output file if it exists
temp_filename
]
result = subprocess.run(ffmpeg_command, capture_output=True, text=True, check=True)
os.replace(temp_filename, filename)
def process_wav_files():
"""Process WAV files: trim silence and transcribe."""
wav_files = [f for f in os.listdir(".") if f.endswith(".wav")]
for wav_file in wav_files:
# Generate the expected transcript filename
transcript_file = os.path.splitext(wav_file)[0] + ".tns"
# Check if transcript already exists
if os.path.exists(transcript_file):
print(f"Transcript already exists for {wav_file}, skipping transcription")
continue
print("Trimming silence: " + wav_file)
trim_silence(wav_file)
with open(wav_file, "rb") as file:
print("Transcribing: " + wav_file)
output_text = whisper_api(file)
output_file = os.path.splitext(wav_file)[0] + ".tns"
with open(output_file, "w") as output:
output.write(output_text)
def chunk_transcript(string, chunk_size):
"""Chunk the transcript to fit in the context window."""
chunks = []
lines = string.split("\n")
current_chunk = ""
for line in lines:
current_chunk += line
if len(current_chunk) >= chunk_size:
chunks.append(current_chunk)
current_chunk = ""
if current_chunk:
chunks.append(current_chunk)
return chunks
def summarize_transcripts():
"""Summarize transcript files."""
today = datetime.datetime.now().strftime('%Y%m%d')
summary_filename = "summary-" + today + ".md"
transcript_files = [f for f in os.listdir(".") if f.endswith(".tns")]
for transcript in transcript_files:
print("Summarizing: " + transcript)
with open(transcript, "r") as file:
transcript_data = file.read()
chunked_data = chunk_transcript(transcript_data, CHUNK_SIZE)
with open(summary_filename, "a") as md_file:
for i, chunk in enumerate(chunked_data):
print("Processing part " + str(i))
summary = llm_local(SUMMARY_PROMPT.format(chunk=chunk))
facts = llm_local(FACT_PROMPT.format(chunk=chunk))
sentiment = llm_local(SENTIMENT_PROMPT.format(chunk=chunk))
md_file.write(f"# Call Transcript - {transcript} - Part {i + 1}\n\nSummary: {summary}\n\nFacts:\n{facts}\n\nSentiment: {sentiment}\n\n---\n")
print("Summarizing complete")
if __name__ == "__main__":
process_wav_files()
summarize_transcripts()