Maya Research released Maya 2 Native, a streaming text-to-speech model. It supports 12 languages including 11 Indian languages and English. The model offers 20 built-in voices and zero-shot voice cloning from short audio. A cloned voice can speak across languages. It generates 24 kHz mono audio in 80 ms streaming frames. The release includes open weights, inference code, and a Python API. It also provides an OpenAI-compatible speech endpoint and WebSocket streaming. Maya 2 Native ranked second on the VoiceArena Hindi TTS leaderboard. The model is licensed under Apache-2.0. It requires Linux, Python 3.10–3.12, and an NVIDIA GPU with CUDA 12.8. A 12 GB or larger GPU is recommended for stable performance.
What changed
Maya 2 Native replaces prior non-streaming or single-language text-to-speech models from Maya Research.
What it unlocks
Streaming text-to-speech in 12 languages with zero-shot cross-lingual voice cloning and self-hosted deployment.
- 12 languages supported
- 20 built-in voices
- 2.5B parameters
- 24 kHz mono audio
What you need to act on it
- Linux OS
- Python 3.10–3.12
- NVIDIA GPU with CUDA 12.8 support
- 12 GB or larger GPU recommended
Sources