Microsoft AI launches streaming transcription model

Microsoft AI announced MAI-Transcribe-2-Streaming, its first streaming transcription model, together with two text-to-speech models: MAI-Voice-2.1 and a faster, cheaper variant, MAI-Voice-2.1-Flash.

According to Microsoft, the transcription model works in 60 languages and detects the spoken language automatically as it goes. The company says it ranks first on Artificial Analysis for accuracy on both final and partial transcripts, and that it sits on the Pareto frontier of that leaderboard's accuracy-versus-latency evaluation. It produces its first partial transcripts, or "partials," just over 100ms after receiving audio, then revises them as more speech arrives. This lets an agent start reasoning or calling tools before the speaker has finished. Microsoft says its internal evaluations show words appearing twice as fast as with its closest competitor, which it does not name. Introductory pricing is $0.54 per hour of audio through the end of the year.

MAI-Voice-2.1 supports 23 languages and 26 locales. A single voice can switch between languages and use a native accent in each. It costs $22 per 1M characters. The Flash variant supports the same languages and is built for high-volume workloads. Microsoft says it can generate 45 seconds of audio with 150ms end-to-end latency, and that it runs inference 55% faster and costs about 60% less than comparable models. It is priced at $15 per 1M characters.

Both voice models can clone a voice in any supported language from a few seconds of reference audio. Microsoft says built-in consent guardrails prevent misuse but did not describe how they work. Apart from the Artificial Analysis ranking, the speed and cost figures come from Microsoft's own testing. A demo agent called Chatter is available in the MAI Playground.