MAI-Voice-2.1

Audio modelCard summarized

Turn text into expressive, natural-sounding speech in seconds

Source: MAI-Voice-2.1 Model Card and MAI-Voice-2.1-Flash Model Card (4 pp)

Use it via API

$0.022 / 1K chars

Code example Compare

Where to call it

Microsoft Foundry / AzureMAI-Voice-2.1
OpenRoutermicrosoft/mai-voice-2.1

Supports

Streaming

Full feature support

Good for

Audiobooks and media narration with a consistent voice
Content creation, voice-over and marketing in one brand voice across markets
E-learning and accessibility narration

Overview

Microsoft · 02Voice

Voice 2.1

Turn text into expressive, natural-sounding speech in seconds

Typical price

$0.022 / 1K chars

MAI-Voice-2.1

Max length: —
Languages: 23 languages, 26 locales
Modes: text-to-speech, instant voice cloning (gated)
Released: 1 Oct 2026

MAI-Voice-2.1 is a prompted text-to-speech model built in-house by Microsoft AI. It speaks 23 languages and 26 locales, can render one voice identity in every supported language, takes emotion and style control through SSML, and clones a voice from 5–60 seconds of audio behind gated, consent-based access. It targets long-form narration and voice-over. MAI-Voice-2.1-Flash is the low-latency sibling for voice agents, with streaming output and a lower price. MAI-Voice-2.1 card p.1; Flash card p.1

MAI-Voice-2.1

Typical price: $0.022 / 1K chars
Languages: 23 languages, 26 locales
Weights: Closed
Released: 1 Oct 2026

text-to-speech
instant voice cloning (gated)
SSML emotion and style control
one voice across all languages
multi-speaker dialogue
long-form generation
streaming (Flash)