The tech giant announced its first streaming transcription model, MAI‑Transcribe‑2‑Streaming, alongside additional voice models that deliver rapid, multilingual captions for conversational AI use cases.

Microsoft unveiled MAI‑Transcribe‑2‑Streaming, its inaugural streaming transcription model designed to provide low‑latency, multilingual real‑time captions for a range of conversational AI applications.

Key capabilities of MAI‑Transcribe‑2‑Streaming

The new model processes audio streams with sub‑second delay, supporting dozens of languages and dialects while maintaining high word‑error rates comparable to batch transcription services. It is optimized for edge deployment, allowing developers to run inference on modest hardware without sacrificing accuracy.

MAI‑Transcribe‑2‑Streaming integrates seamlessly with Microsoft’s existing Azure Speech services, enabling developers to add live captioning to virtual assistants, live‑streamed events, and interactive voice‑driven platforms.

Additional voice models released

Alongside the streaming model, Microsoft introduced two new voice‑generation models: MAI‑Voice‑2‑Neural, which delivers more natural prosody for text‑to‑speech, and MAI‑Voice‑2‑Low‑Latency, tuned for rapid response in dialogue systems.

Both models are built on the same underlying architecture as MAI‑Transcribe‑2‑Streaming, ensuring consistent performance across speech‑to‑text and text‑to‑speech pipelines.

Potential use cases

  • Live captioning for webinars and virtual conferences
  • Real‑time subtitles for video‑calling platforms
  • Interactive voice assistants that require instant feedback
  • Multilingual customer‑service bots handling simultaneous conversations

Microsoft highlighted early adopters in the education sector, where instant multilingual subtitles improve accessibility for remote learners, and in contact‑center environments, where agents can receive live transcription of customer speech to enhance response accuracy.

Our goal is to make real‑time speech understanding universally available, regardless of language or device constraints.

The rollout is initially available in preview on Azure, with broader public availability planned later this year as Microsoft continues to expand its AI‑first strategy.

For full details, see the Microsoft AI announcement: Microsoft AI coverage of MAI‑Transcribe‑2‑Streaming launch.