Models · 1 October 2026 · 1 min read
MAI-Voice-2.1: one voice in 23 languages
Released on 1 October 2026: text-to-speech in 23 languages and 26 locales, with the same voice speaking every one of them.
By The Crewnema team
Microsoft AI released two text-to-speech models on 1 October 2026: MAI-Voice-2.1, and MAI-Voice-2.1-Flash for fast, high-volume work. These are the points from Microsoft’s announcement.
What Microsoft announced
- 23 languages and 26 locales, with one voice able to speak all of them with a native accent.
- Voice cloning in every supported language from a few seconds of reference audio, with consent guardrails built in.
- MAI-Voice-2.1 at $22 per million characters; MAI-Voice-2.1-Flash at $15, generating 45 seconds of audio with 150 ms end-to-end latency, by Microsoft’s figures.
- Availability on Microsoft Foundry, OpenRouter, MAI Playground, Vercel and Azure Voice Live.
Our view
Of this month’s voice releases, this is the one built for what advertisers do. One voice across 23 languages means the same brand voice can read an advert in English and again in another language, which is the cheapest way to sell in more than one market. The price is ordinary for the category. What we would check first is which Indian languages are among the 23, and how a cloned voice handles names and prices; Microsoft’s list did not answer the first in what we read.
On cloning: use your own voice or one you have written permission for. Consent guardrails help, but they are not your permission.
Side by side
In Crewnema
MAI-Voice-2.1 is not on Crewnema’s roster; voiceovers use the voice models listed in how the levels work, and you can attach your own recording to any shot.
Sources
Make your first video
Crewnema is in private beta and lets businesses in a few at a time. Request access and we will write when yours is open.