Skip to content

Latest commit

 

History

History
37 lines (29 loc) · 1.37 KB

training.md

File metadata and controls

37 lines (29 loc) · 1.37 KB

Training

Before training, please install MeloTTS in dev mode and go to the melo folder.

pip install -e .
cd melo

Data Preparation

To train a TTS model, we need to prepare the audio files and a metadata file. We recommend using 44100Hz audio files and the metadata file should have the following format:

path/to/audio_001.wav |<speaker_name>|<language_code>|<text_001>
path/to/audio_002.wav |<speaker_name>|<language_code>|<text_002>

The transcribed text can be obtained by ASR model, (e.g., whisper). An example metadata can be found in data/example/metadata.list

We can then run the preprocessing code:

python preprocess_text.py --metadata data/example/metadata.list 

A config file data/example/config.json will be generated. Feel free to edit some hyper-parameters in that config file (for example, you may decrease the batch size if you have encountered the CUDA out-of-memory issue).

Training

The training can be launched by:

bash train.sh <path/to/config.json> <num_of_gpus>

We have found for some machine the training will sometimes crash due to an issue of gloo. Therefore, we add an auto-resume wrapper in the train.sh.

Inference

Simply run:

python infer.py --text "<some text here>" -m /path/to/checkpoint/G_<iter>.pth -o <output_dir>