Run Maya 2 Native locally: streaming TTS in 12 languages
We’re releasing Maya 2 Native , a text-to-speech model from Maya Research for 11 Indic languages and English . The weights and inference code are available under Apache-2.0 , with 20 built-in voices , streaming audio, and voice cloning from 3–10 seconds of reference speech . Get the weights · Inference code · Full model card and evaluations Watch the launch video Hear it first The model supports…
Maya Research has launched Maya 2 Native, a text-to-speech model supporting 11 Indic languages and English. The open-source weights and inference code are available under Apache-2.0, with 20 voices, streaming audio, and voice cloning from short reference recordings. Key features include 24 kHz mono audio, streaming capabilities, and support for voice cloning with 3-10 seconds of input.
The model can be installed using Python 3.10-3.12 and an NVIDIA GPU with CUDA 12.8. A Python API, OpenAI-compatible endpoint, and WebSocket streaming are provided, with steady-state latency of 111-242 ms and throughput of 323 seconds of audio per second on a H100 NVL 224. Evaluation shows 3.49% Character Error Rate across six thousand held-out texts, with good voice identity retention.
Assamese remains a challenging language. Feedback from developers is welcomed to prioritize improvements in future releases.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
