MAI-Transcribe-2-Streaming
Microsoft AI’s first streaming speech-to-text model, in 60 languages.
MAI-Transcribe-2-Streaming transcribes live audio in 60 languages with automatic, continuous language detection. Microsoft says it emits first partial text just over 100 ms into the audio, then revises it and commits stable text. It is offered in Microsoft Foundry, MAI Playground and Azure Voice Live.
- Provider
- Microsoft AI
- Type
- Speech-to-text
- Released
- Oct 1, 2026
- Input
- audio
- Output
- text
- Open weights
- No
Best for
- Real-time / low latency
- Extraction & classification
- Enterprise
Strengths
- ≈100 ms to first partial text (vendor-reported)
- 60 languages, auto-detected
More from Microsoft AI
- Microsoft-Decision-1 — Microsoft AI. Microsoft’s small decision-scoring model: calibrated probabilities over fixed options.
- MAI-Image-2.6 — Microsoft AI. Microsoft AI’s in-house image model.
- MAI-Image-2.7 Preview — Microsoft AI. Preview of Microsoft AI’s next image model, editing only for now.
- MAI-Voice-2.1 — Microsoft AI. Microsoft AI’s most expressive multilingual text-to-speech model.
- MAI-Voice-2.1-Flash — Microsoft AI. Low-latency MAI-Voice for real-time, high-volume speech.
- MAI-Transcribe-2 — Microsoft AI. Microsoft AI’s in-house speech-to-text model.