vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2025-09-15 11:29:43 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

wip, adding IPA for kokoro

Browse source
This commit is contained in:
Concedo 2025-08-18 00:51:12 +08:00
parent 3138a151c2
commit 3f621be7dd
3 changed files with 65185 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

65118
kokoro_ipa.embd Normal file
View file

File diff suppressed because it is too large Load diff

21
otherarch/ttscpp/src/kokoro_model.cpp
View file

@ -1387,6 +1387,24 @@ std::vector<std::vector<uint32_t>> kokoro_runner::tokenize_chunks(std::vector<st
return chunks;
}
//kcpp hacked a quick replace fn
static void kokoro_str_replace_all(std::string & s, const std::string & search, const std::string & replace) {
if (search.empty()) {
return;
}
std::string builder;
builder.reserve(s.length());
size_t pos = 0;
size_t last_pos = 0;
while ((pos = s.find(search, last_pos)) != std::string::npos) {
builder.append(s, last_pos, pos - last_pos);
builder.append(replace);
last_pos = pos + search.length();
}
builder.append(s, last_pos, std::string::npos);
s = std::move(builder);
}
int kokoro_runner::generate(std::string prompt, struct tts_response * response, std::string voice, std::string voice_code) {
if (model->voices.find(voice) == model->voices.end()) {
fprintf(stdout,"\nFailed to find Kokoro voice '%s' aborting.\n", voice.c_str());
@ -1406,7 +1424,10 @@ int kokoro_runner::generate(std::string prompt, struct tts_response * response,
// We preserve the other punctuation for cleaner chunking pre-tokenization
prompt = replace_any(prompt, ",;:", "--");
prompt = replace_any(prompt, "\n", " ");
kokoro_str_replace_all(prompt," - "," -- ");
kokoro_str_replace_all(prompt,"'s ","s ");
std::string phonemized_prompt = phmzr->text_to_phonemes(prompt);
// printf("\nRESULT: %s\n",phonemized_prompt.c_str());
// Kokoro users a utf-8 single character tokenizer so if the size of the prompt is smaller than the max context length without the
// beginning of sentence and end of sentence tokens then we can compute it all at once.

46
otherarch/ttscpp/src/phonemizer.cpp
View file

@ -798,9 +798,53 @@ bool phonemizer::handle_phonetic(corpus* text, std::string word, std::string* ou
return true;
}
static std::unordered_map<std::string, std::string> kokoro_ipa_map;
void populate_kokoro_ipa_map(std::string executable_path)
{
std::string line;
auto filepath = executable_path + "kokoro_ipa.embd";
printf("\nReading Kokoro IPA from %s",filepath.c_str());
std::ifstream myfile(filepath);
if (myfile.is_open())
{
while (myfile.good())
{
getline(myfile, line);
auto parts = split(line, ",");
if(parts.size()==2)
{
kokoro_ipa_map[parts[0]] = parts[1];
} else {
printf("\nError reading line in Kokoro IPA!");
}
}
myfile.close();
}
else
{
printf("\nUnable to open Kokoro IPA file");
}
}
std::string found_word_to_ipa(std::string input)
{
auto it = kokoro_ipa_map.find(input);
if (it != kokoro_ipa_map.end()) {
return it->second; //found
}
return "";
}
bool phonemizer::process_word(corpus* text, std::string* output, std::string word, conditions* flags, bool has_accent) {
dictionary_response* response;
size_t unaccented_size_difference = 0;
std::string foundstr = found_word_to_ipa(word);
if(foundstr!="")
{
output->append(foundstr);
text->size_pop(word.size());
return true;
}
if (has_accent) {
response = dict->lookup(text, word, flags);
if (!response->is_successful()) {
@ -813,6 +857,8 @@ bool phonemizer::process_word(corpus* text, std::string* output, std::string wor
response = dict->lookup(text, word, flags);
}
//printf("\nSUCCESS: %d, word:%s, result:%s\n",response->is_successful(),word.c_str(),response->value.c_str());
if (response->is_successful()) {
if (flags->was_word && output->back() != ' ' && !flags->hyphenated) {
output->append(" ");