Microsoft AI introduced MAI-Voice-2.1 in early October 2026, a text-to-speech model aimed at long-form content such as audiobooks and voice-overs. If you turn ebooks, courses or lessons into audio products, it is worth knowing what it offers, what it costs and where the limits are.
What Microsoft says it does
Microsoft lists two models. MAI-Voice-2.1 is the higher-fidelity version, recommended for audiobooks, content creation and voice-over. MAI-Voice-2.1-Flash has lower latency and targets call-center agents and voice assistants, so it is less relevant to product creators. Both support 23 languages and emotion control. Microsoft’s documentation says the main model keeps a speaker consistent across long passages.
What it costs
Microsoft’s model page lists MAI-Voice-2.1 at $22 per million characters, and the Flash version at $15 per million. As a rough estimate, a 60,000-word ebook is around 360,000 characters at about six characters per word, which works out to roughly $8 at the listed rate. That is our own arithmetic, not a Microsoft figure, and it ignores retries and editing passes.
Where you can use it
The model page names the MAI Playground, Copilot Audio Expressions and Microsoft Foundry (Azure Speech). Microsoft’s Learn documentation says both models are in public preview, with no service-level agreement, and are not recommended for production workloads. Treat it as something to test, not a tool to build a business on yet.
Voice cloning is gated
Microsoft describes instant voice matching from a short reference clip, but the documentation says production use requires an approved, consented voice and an access review. Unlicensed cloning is not possible. For product sellers, that is a useful reminder: never clone a real person’s voice without clear written permission.
Before you sell the audio
- Check the rights to the text. Not every PLR license allows audio versions or derivatives. Read yours first.
- Read the output terms. Microsoft’s documentation says Microsoft holds full licensing rights for commercial use. Confirm what that means for you in the service terms before publishing.
- Check the platform’s rules. Audiobook stores may require disclosure of AI narration. See our guide to AI-narrated audiobooks.
- Listen to the whole file. Names, numbers and unusual words are common trouble spots for any synthetic voice.
Microsoft’s pages do not mention watermarking or maximum text length, so check the documentation before planning a large batch.
Sources
- Microsoft AI: MAI-Voice-2.1
- Microsoft Learn: MAI-Voice-2.1 and MAI-Voice-2.1-Flash
- Digit: Microsoft launches MAI-Transcribe-2-Streaming and MAI-Voice-2.1
Featured image: Photo: Wikimedia Commons (CC0)