A meeting assistant must show a live transcript that indicates which participant said…

How it works

Diarization tells apart the different speakers in a conversation. The Speech SDK's ConversationTranscriber does speech to text with real-time diarization: each result says who was speaking in a speaker ID field. That ID is a generic identifier the service assigns to each participant as it picks out different speakers in the audio, so you see values like Guest-1 and Guest-2, not people's names.

Real-time transcription transcribes live audio instantly and returns intermediate results, which is what a live meeting view needs. The quickstart sample turns on the SpeechServiceResponse_DiarizeIntermediateResults property. In its output, the first intermediate (TRANSCRIBING) results show Speaker ID=Unknown, and later ones and the final TRANSCRIBED result show Guest-1. Show that progression in your UI instead of assuming every partial result already has a speaker. The service can identify up to 35 different speakers in a recording.

For a recording you already have, rather than a live meeting, the page also points to the fast transcription API. It returns results synchronously, faster than real time, and supports diarization.

The other options, and when you'd use them

• Batch transcription with diarization, polled during the meeting: Batch transcription handles large amounts of audio in storage. You submit audio and retrieve the results asynchronously. Jobs are scheduled on a best-effort basis, and at peak hours a job can take up to 30 minutes to start and up to 24 hours to finish. Use it for large volumes of prerecorded audio, such as analyzing recorded call center calls.
• Speech to text REST API for short audio, with diarization: This API doesn't support real-time diarization. It also returns only final results with no partial results, and requests that send audio directly can hold no more than 60 seconds. Use it only when you can't use the Speech SDK or the fast transcription API.
• ConversationTranscriber returning real names from Microsoft Entra: The speaker ID is a generic identifier the service assigns, not a name from a directory profile.

Exam tip

"Live transcript and who said what" means ConversationTranscriber with real-time diarization and generic Guest-n IDs. "Thousands of stored files" means batch transcription, and "one recording, fast" means fast transcription.

References

• Quickstart: Create real-time diarization
• What is speech to text?
• What is batch transcription?
• Use speech to text REST API for short audio