#1 Real-time · Hindi
Maya 2 Native
Our flagship model. Built natively in each language, so it keeps the accent and rhythm of a real speaker.
#2 overall · Elo 1086 · Voice Arena
See it on Voice Arena
Seed $1.9M from South Park Commons
Voice models that sound native
Real-time speech that carries the accent, rhythm, and emotion of a native speaker.
Listen
Hear the difference
Pick a language, pick a context, and press play. Every sample is real Maya audio.
बहुत बहुत धन्यवाद! आपकी मदद के बिना ये मुमकिन नहीं था।
Warmth, urgency, apology. The tone follows the meaning of the words.
Speech that starts instantly
Audio begins before the model finishes generating. A long paragraph starts speaking just as fast as a single line.
75ms
Time to first byte
7.3ms
Variation between calls
#1
Real-time Hindi · Voice Arena
10+
Languages supported
Measured from Bangalore to our Mumbai region.
Three models hosted, two fully open
#3 Real-time · English
Maya 2 Global
One model for every market. It holds its own on the English board against labs many times our size.
#4 overall · Elo 1024 · Voice Arena
See it on Voice Arena 1089 Elo score
Maya 2 Flash
The fastest model we make. Built for live conversation, where every millisecond counts.
Artificial Analysis leaderboard
See it on Artificial AnalysisOpen weights
Maya 1
Apache 2.0. Download it, fine-tune it, ship it in your own stack.
Veena
Our first Hindi and English model. Still open, still downloaded every day.
The whole API
One endpoint.
Send text, choose a voice, read the stream. No sessions, no job queues, no files to download afterwards.
- Regions
- Mumbai · Iowa
- Output
- 24 kHz PCM, streamed
curl
curl -X POST https://tts.mayaresearch.ai/v1/tts \
-H "Authorization: Bearer $MAYA_API_KEY" \
-H "content-type: application/json" \
-d '{"voice":"Ananya",
"text":"नमस्ते! आपका ऑर्डर कल पहुँच जाएगा।",
"language":"hi"}' \
--output out.pcm