Model credits
The Bengali voice was trained on synthetic Bengali speech with the Tiny Voca architecture (Apache 2.0).
- Maya 2 Native generated the Jahnavi training audio (Apache 2.0). Maya uses Kyutai Mimi for audio decoding (CC BY 4.0).
- Training prompts came from Mozilla Common Voice's Bengali Sentence Collector (CC0 1.0).
- English pronunciation fallback uses the CMU Pronouncing Dictionary (BSD-style license).
- The speech engine runs with ONNX Runtime Web (MIT).
- The Bengali interface uses Noto Sans Bengali (SIL Open Font License).