Add support for VITS (multilingual TTS) #466
Add custom VITS tokenizer converter
6c7ed967
Do not decode if expected input_ids is empty
53e78a03
Update vits tokenizer tests
21ed1f1c
Implement `VitsTokenizer`
aa66b4f7
Add support for VITS model
f1ca1d90
Support VITS through pipeline API
d78cf16a
Update JSDoc
019ed2b5
Add TTS unit test
381b5eb1
Add speecht5 unit test
5da17e0a
Fix typo
df39fa4e
Fix typo
83d33720
Update speecht5 model id
27873298
Add note about using quantized speecht5 in unit tests
9f0ccfeb
Monkey-patch `BigInt64Array` and `BigUint64Array`
fb4850a6
Merge branch 'main' into add-vits
b4a78bfb
xenova
merged
1394f731
into main 1 year ago
xenova
deleted the add-vits branch 42 days ago
Assignees
No one assigned
This PR adds support for VITS (text-to-speech). In particular, it allows us to run Meta's large collection of MMS models, covering over 1000 different languages!
closes #429
Example usage:
English:
english.mp4
French:
french.mp4
Surprisingly, the quantized versions of the models still work quite well 👍.
See here for the full list of supported languages.