Breaking

Microsoft launches expressive TTS model MAI-Voice-2 with 15 languages

June 23, 2026 at 12:30 EDT

Media Generation
Foundation Models
Sales & Business

Microsoft on June 2, 2026 released MAI-Voice-2, a new text-to-speech model that generates natural, expressive speech from text or a short reference clip. It supports 15 languages and is available through Microsoft Foundry and Azure Speech.

Build 2026 · Microsoft AI

MAI-Voice-2: high-fidelity speech that speaks with emotion

Microsoft's new multilingual text-to-speech model expresses tone, emotion and delivery across 15 languages — and reached speaker similarity that's hard to tell from a human recording.

languages & locales supported

72%

preferred over MAI-Voice-1 in side-by-side tests

$22

per 1M characters, starting price

Listener preference, head to head

Share of side-by-side comparisons each model won

72%

MAI-Voice-2

28%

MAI-Voice-1

What it can do

Emotion control

Sad, whispered, excited, anger and intensity via the mstts:express-as tag.

Code-switching

Blends languages mid-sentence, e.g. Hindi-English or Spanish-English.

Voice prompting

Zero-shot voice from a 5-60s clip — gated, consent-based, with guardrails.

Well regarded

Expressive emotion control
Long-form speaker consistency
Natural multilingual speech

Open constraints

Version drift & at-scale validation
Voice cloning gated, not open
Low-latency Flash variant still pending

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.