Training Guide

Use Python 3.9, due to a bug in fairseq. Install all dependencies with pip install 'vocos[train]'.

First training, it will automatically download models from Huggingface. See the file metrics/UTMOS.py.

Please resample the audio files prior to training.

Run python -m genlist to generate train / valid file list.

Run python train.py -c configs/vocos.yaml to train the model.