New TTS model converted from the Pocket TTS
- simpler chunking algorithm while supporting longer token counts
- EOS detection stops generation naturally instead of hitting a fixed token wall
- real time support
- No espeak dependency
- iOS RAM friendly
Bug Fixes
- #282 — Custom vocabulary rescoring now applies to chunked long audio (was silently skipped on >15s
audio) — thanks @Beingpax
Note
- we plan to deprecate kokoro tts in the future soon
Contributors
Beingpax