|
--- |
|
language: |
|
- hi |
|
- en |
|
library_name: transformers |
|
pipeline_tag: text-to-speech |
|
--- |
|
XTTSv2 checkpoints finedtuned with the forked Coqui TTS (https://github.com/idiap/coqui-ai-TTS) for Hindi speech |
|
|
|
Trained using the Indic TTS Database (https://www.iitm.ac.in/donlab/tts/) and Mozilla Common Voice 18.0 Hindi dataset (https://commonvoice.mozilla.org/en/datasets) |
|
|
|
Rename checkpoint to model.pth and replace original XTTSv2 model, or use according to how your implementation of XTTSv2. |
|
|
|
The checkpoints with the highest step count may _not be the best_. I think the best quality output here is best_model_43036.pth Reference speaker audio files are in ./speakers-hi of this repo |
|
|
|
Use language code 'hi' at inference for Hindi speech, use language code 'hi' with English text to generate English with the learned Hindi pronounciations. |