Speaker Attributed ASR

Speaker Attributed ASR (SAA) is a rich transcription mode that asks Granite Speech to produce a transcript with speaker-turn labels. Use it when you want a speaker-attributed text transcript, not acoustic diarization segments. For “who spoke when” timelines with timestamps, use Diarization.

When to Use SAA

Use SAA when…	Use diarization when…
You want a readable `[Speaker N]:` style transcript.	You need speaker segments with start/end times.
Granite Speech’s language model can infer speaker turns from the audio.	You need acoustic speaker separation from the Sortformer pipeline.
You do not need streaming or timestamp alignment.	You need reruns, alignment metrics, and diarization-specific quality controls.

Model Requirement

SAA currently requires:

izwi pull Granite-Speech-4.1-2B-Plus

If the API request omits model_id, the persisted SAA workflow defaults to Granite-Speech-4.1-2B-Plus. Supplying a different model returns a validation error.

Using the Web UI

Open Transcription in the sidebar.
Choose Speaker Attributed ASR from the mode switch.
Upload or record audio.
Select a ready Granite Speech model.
Choose a speaker expectation: Auto, 2+, 3+, or 4+.
Optionally enable summary generation.
Submit the job and review the speaker-turn transcript.

SAA disables streaming and timestamp alignment. Those controls belong to normal transcription or diarization workflows.

Using the API

Create a persisted SAA job with job_kind=speaker_attributed_asr:

curl -X POST "http://localhost:8080/v1/speech-to-text/jobs?job_kind=speaker_attributed_asr" \
  -F "file=@meeting.wav" \
  -F "model_id=Granite-Speech-4.1-2B-Plus" \
  -F "language=English" \
  -F "generate_summary=true" \
  -F "min_speakers=2"

The short alias job_kind=saa is also accepted. Poll the returned record until processing_status is ready:

curl "http://localhost:8080/v1/speech-to-text/jobs/saa_123?job_kind=speaker_attributed_asr"

Fetch stored audio or regenerate summaries with the same job kind:

curl "http://localhost:8080/v1/speech-to-text/jobs/saa_123/audio?job_kind=speaker_attributed_asr" \
  --output source.wav

curl -X POST \
  "http://localhost:8080/v1/speech-to-text/jobs/saa_123/summary/regenerate?job_kind=speaker_attributed_asr"

Request Fields

JSON and multipart create requests accept:

Field	Description
`file` / `audio_base64`	Source audio upload.
`model_id` / `model`	Optional model override. Must be `Granite-Speech-4.1-2B-Plus` when present.
`language`	Optional language hint, such as `English`.
`generate_summary`	Generate an AI summary after the transcript completes. Defaults to `false`.
`min_speakers`, `max_speakers`	Optional speaker expectation bounds. `min_speakers` is what the current UI sends.

Unsupported for SAA:

stream
include_timestamps
word_timestamps
aligner_model_id

Home

Reference & Help

Speaker Attributed ASR

When to Use SAA

Model Requirement

Using the Web UI

Using the API

Request Fields

See Also

​When to Use SAA

​Model Requirement

​Using the Web UI

​Using the API

​Request Fields

​See Also

When to Use SAA

Model Requirement

Using the Web UI

Using the API

Request Fields

See Also