Maya Research releases Maya 2 Native TTS model for 12 languages

Maya Research has open-sourced Maya 2 Native under Apache-2.0, with weights and inference code. The text-to-speech model covers 11 Indic languages plus English, offers 20 built-in voices, streaming audio and voice cloning from 3–10 seconds of reference speech.

• Model outputs 24 kHz mono audio and streams over WebSocket
• Serves 224 concurrent streams on a single H100 NVL
• Median time to first audio is 111 ms
• Average CER on a 6,000-text benchmark is 2.86–3.49%