abylouw / convnext_tts

Unofficial implementation of ConvNeXt-TTS powered by lightning and Rye

convnext-tts

Unofficial implementation of ConvNeXt-TTS(paper) for my experiment. The model architecture has been slightly modified.

Usage

0. Install dependencies using Rye(link).
1. Download JSUT corpus and fullcontext label(link) and then sample wave files(basic5000) to 24kHz.
2. Create a default.yaml file under the convnext_tts/bin/conf/pathdirectory, setting wav_dir, lab_dir and data_root according to your environment, using src/convnext_tts/bin/conf/path/dummy.yaml as a reference.
3. Run exp/jsut/run.sh.

Now I'm running it, but it seems likely to fail....

Still under development...