This extension perform voice conversion of audio file sent by ST using (a monkey patched) RVC.
Features
a small script getting TTS audio from request, applying requested RVC model on it and sending back the new audio
What changed
Needed to correct coqui extension scripts to keep current directory path
RVC and fairseq package dirty patched version for now in modules/voice_conversion/rvc and modules/voice_conversion/fairseq
Limitations
Need the RVC models to be stored in data/models/rvc (index file optional)
RVC code is a strip version from original repository with only what is needed for performing voice changing
RVC code relies onf fairseq that is not compatible with python 3.11, made a monkey patched version of fairseq that run fine in module modules\voice_conversion\fairseq
Their is a theorytical dependency conflict between TTS and Torchcrepe over librosa, cannot install everything with one requirement file, made an additional requirement file to be call after for now
Testing
The extension works fine alongside coqui when installed as follow:
conda install python=3.11
pip install -r requirements-complete.txt
pip install -r requirements-rvc.txt
put RVC models into "data/models/rvc/" index files are optional
example extras call: python server.py --enable-modules=rvc,live2d,classify,coqui-tts --coqui-gpu
For ST:
RVC voice map example:
Seraphina=guanguanV1, Coding Sensei=kiki
This extension perform voice conversion of audio file sent by ST using (a monkey patched) RVC.
Features
What changed
Limitations
Testing
The extension works fine alongside coqui when installed as follow:
For ST: RVC voice map example: Seraphina=guanguanV1, Coding Sensei=kiki
Coqui voice map example: Seraphina:tts_models--en--ljspeech--glow-tts\model_file.pth, Coding Sensei:tts_models--multilingual--multi-dataset--your_tts\model_file.pth[3][0]