Wordloop09 Speaker Labelling
Speaker labels can be assigned to people and survive final transcript replacement.
09 Speaker Labelling
Goal
A user can assign transcript speaker labels to people, create missing people, and keep those assignments when the final transcript replaces provisional transcript output.
Why This Comes Now
Speaker labels depend on transcript identity and final replacement semantics. They come after task reconciliation because both require user-owned metadata to survive regenerated ML output without being overwritten.
Included Scope
- Show speaker labels on transcript turns or segments.
- Assign an existing person to a speaker label.
- Create a new person from the speaker assignment flow.
- Persist speaker-to-person assignments on the meeting.
- Preserve assignments when the final transcript replaces provisional live transcript output.
Not Included
- Automatic speaker identity recognition across meetings.
- Speaker-aware task assignment or summary rewriting.
- Contact enrichment, avatars, or organization directory sync.
- Audio playback review and synced transcript highlighting.
Domain Slices
| # | Domain | Slice summary | Scaffold command | Complexity | Prerequisite |
|---|---|---|---|---|---|
| 1 | Core | POST /meetings/{id}/speaker-labels; person lookup / create; speaker_label → person_id mapping persisted to transcript_segments; mapping survives PUT /transcriptions/{id}/segments replacement; SpeakerStateUpdatedEvent to ML during live session | ./dev new slice meeting-recording 09-speaker-labelling core speaker-assignment | M | Milestone 04 merged |
| 2 | App | Speaker label display on transcript segments; inline assignment dropdown (existing person or create new); assignment optimistically applied; unassigned labels visually distinct | ./dev new slice meeting-recording 09-speaker-labelling app speaker-labelling-ui | M | Slice 1 merged |
| 3 | ML | speaker_label field populated on TranscriptSegmentProducedEvent from AssemblyAI diarization; SegmentFeaturesProducedEvent for voice embeddings | ./dev new slice meeting-recording 09-speaker-labelling ml speaker-diarization | S | Milestone 04 merged |
Acceptance
These criteria map to tests/bets/meeting-recording/test_milestone_09_speaker_labelling.py. Run via ./dev test bet meeting-recording.
- Transcript output displays distinct speaker labels.
- User can assign an existing person to a speaker label.
- User can create a person from the speaker labelling flow.
- Speaker assignments persist after refresh.
- Stopping and finalizing the recording replaces provisional transcript output without dropping assigned people.
- Unassigned speaker labels remain clearly visible for later review.
ML — Live Task Extraction
TaskProducedEvent emission from the same batched insight pipeline as talking points, and final task candidates in the PUT /meetings/:id/tasks/system write-back.
App — Speaker Labelling UI
Speaker label display on transcript segments, inline assignment dropdown (existing person or create new), optimistic assignment, and visually distinct unassigned labels.