Microsoft AI launches MAI-Transcribe-2-Streaming and MAI-Voice-2.1 models

On 1 October 2026, Microsoft AI launched MAI-Transcribe-2-Streaming, its first streaming transcription model, together with text-to-speech models MAI-Voice-2.1 and MAI-Voice-2.1-Flash, available through Microsoft Foundry and related channels, Unite.AI and Microsoft communications reported.

Streaming transcription claims

Microsoft describes MAI-Transcribe-2-Streaming as low-latency real-time transcription in 60 languages with continuous automatic language detection. Citing the Artificial Analysis streaming leaderboard dated 28 September 2026, company materials highlighted roughly a 2.5% final word-error rate, 2.8% first-partial rate and about 0.13 seconds to final transcription, with first partials said to appear just over 100 milliseconds after audio arrives. Introductory pricing was listed at $0.54 per hour of audio through year-end.

Voice models

MAI-Voice-2.1 supports 23 languages / 26 locales, with a single voice identity across languages (priced at $22 per 1M characters). MAI-Voice-2.1-Flash targets high-volume, low-latency use — up to 45 seconds of audio with about 150 ms end-to-end latency — at $15 per 1M characters, with Microsoft claiming roughly 55% faster inference and about 60% lower cost versus comparable models. Both support few-second voice cloning with consent guardrails, Microsoft said. A 4,000-listener Turing-style test was cited in which 50.3% of listeners rated MAI-Voice as equally or more human-like than human recordings.

Availability and caveats

Models are available via Microsoft Foundry, the MAI Playground, Vercel and Azure Voice Live (LiveKit listed as coming soon); voice models also via OpenRouter. Benchmark leadership and human-likeness figures are vendor-cited evaluations — readers should treat Artificial Analysis and internal Microsoft tests as reported claims, not independent verification by this site. This is a product launch summary, not an endorsement.