EveryVoiceTTS / EveryVoiceTTS/EveryVoice
[FastSpeech2] Investigate fine-tuning of models with different pitch/energy bins
Open
enhancement
help wanted
- Dominant language
- Python
- Stars
- 45
- Forks
- 4
- Avg merge
- 1d 2h
- Merged PRs (30d)
- 14
Description
The pitch and energy bins are determined based on the `stats.json` of a particular dataset, which will change during fine-tuning. Currently the bins will change and the model has to adapt and learn how they have changed, but we should maybe consider some other fine-tuning strategies and pay attention to how energy/pitch losses perform in different pretrain/finetuning set-ups
Contributor guide
Assessment
This issue has not been assessed yet.