$4 and 103 Milliseconds
Maya 2 Native is the cheapest voice on our text-to-speech board, and it answered fastest of the seven systems we timed on Hindi text. Two axes, and neither of them is how natural it sounds.
We timed seven text-to-speech systems on Hindi text. Maya 2 Native came back first, by a wide margin.
Hindi, time to first audio
Thirty calls per system from us-east4, concurrency of one, warmup discarded. Deepgram aura-2 ran as the control and reproduced its published cell, 123ms against 125ms. Maya got two sweeps and the rest got one, so read their p90s loosely.
| System | p50 | p90 |
|---|---|---|
| Maya 2 Native | 103ms | 111ms |
| Cartesia | 129ms | 231ms |
| Google Chirp 3 HD | 173ms | 312ms |
| MiniMax | 317ms | 341ms |
| Gemini | 493ms | 642ms |
| ElevenLabs | 639ms | 983ms |
| Speechify | 826ms | 956ms |
Maya’s English runs at the same speed as its Hindi, inside our noise, across both sweeps. We did not time an English baseline for the other six, so this table says which system answers fastest on Hindi text and nothing about what any language costs anyone.
It is also the cheapest row we publish
$4 per million characters, against $10 for the next two rows on our English board, where it is second fastest of twenty.
Maya publishes no rate card. That $4 came from them directly, so it is the one price on the board you cannot go and check, and it carries a tilde.
Before you pin it
Time to first audio is a transport number. It says nothing about how well a system reads the language, and we have not measured that here.
Naturalness is a different axis, and Maya is mid-field on it in Hindi, Tamil and Telugu. Those columns are on the multilingual board.
Maya covers eleven languages and all of them are Indic, so its English is Indian English with no American or British variant. A neutral-English route must not land here.
Every number above is measured from us-east4. A caller dialling from India is a different route and we have not timed it.