MAI-Transcribe in Azure Speech (preview)

To start using transcription with enhanced mode, first follow the LLM Speech quickstart. Then, specify the Model.

To use the MAI-Transcribe model:

• Set the enhancedMode.enabled property to true in the request.
• Set the enhancedMode.model property to "MAI-Transcribe-2" in the request.
• Use one of the regions where MAI-Transcribe is available Speech service regions.

curl --location 'https://YourResourceName.cognitiveservices.azure.com/speechtotext/transcriptions:transcribe?api-version=2025-10-15' \
--header 'Content-Type: multipart/form-data' \
--header 'Ocp-Apim-Subscription-Key: <YourSpeechResourceKey>' \
--form 'audio=@"YourAudioFile.wav"' \
--form 'definition={ 
  "enhancedMode": {
    "enabled": true,
    "model": "MAI-Transcribe-2"
  }
}'

Optionally, for MAI-Transcribe-2, enable speaker diarization to distinguish between speakers and attribute words to the right person within a recording.

 --form 'definition={ 
  "enhancedMode": {
    "enabled": true,
    "model": "MAI-Transcribe-2"
  },
  "diarization": {"enabled": true}
}

Optionally, for MAI-Transcribe-2, control the transcript style using transcribeStyle. The default output is verbatim if you need the raw spoken content preserved, including fillers (“um”, “uh”), false starts, and self‑corrections. Set transcribeStyle to clean if you need a readability‑optimized transcript that removes filler words and auto‑formats common speech patterns.

 --form 'definition={ 
  "enhancedMode": {
    "enabled": true,
    "model": "MAI-Transcribe-2",
    "modelOptions": {
      "transcribeStyle": "clean"
    }
  }
}

Optionally, for MAI-Transcribe-2, enable word‑level timestamps to provide precise timing for every word, supporting accurate alignment, search, navigation, and editing.

 --form 'definition={ 
  "enhancedMode": {
    "enabled": true,
    "model": "MAI-Transcribe-2",
    "modelOptions": {
      "timestamps": "word"
    }
  }
}

Optionally, add a list of phrases to increase accuracy in specialized domains by using phraseList.phrases. This feature implements keyword biasing.

 --form 'definition={ 
  "enhancedMode": {
    "enabled": true,
    "model": "MAI-Transcribe-2"
  }, 
  "phraseList": {
     "phrases": ["phrase 1", "phrase 2"]
   },
 }'

Optionally, specify a language code in locales to force recognition in a single language. If you don't specify, the model automatically detects the language. For example:

--form 'definition={
  "locales": ["en"],
  "enhancedMode": {
    "enabled": true,
    "model":"MAI-Transcribe-2"
  }
}'

To start using the MAI-Transcribe model, first follow the Azure Speech API's quickstart guide to use Fast Transcription API's enhancedMode. Then, specify MAI-Transcribe-2 in the enhancedMode.model property. See the complete guide for all the parameters supported by MAI-Transcribe, currently only available under the REST API tab.

To start using the MAI-Transcribe model, first follow the Azure Speech API's quickstart guide to use Fast Transcription API's enhancedMode. Then, specify MAI-Transcribe-2 in the enhancedMode.model property. See the complete guide for all the parameters supported by MAI-Transcribe, currently only available under the REST API tab.

To start using the MAI-Transcribe model, first follow the Azure Speech API's quickstart guide to use Fast Transcription API's enhancedMode. Then, specify MAI-Transcribe-2 in the enhancedMode.model property. See the complete guide for all the parameters supported by MAI-Transcribe, currently only available under the REST API tab.

To start using the MAI-Transcribe model, first follow the Azure Speech API's quickstart guide to use Fast Transcription API's enhancedMode. Then, specify MAI-Transcribe-2 in the enhancedMode.model property. For a complete guide to all the parameters supported by MAI-Transcribe, see the REST API tab.