Research: Speaker identification and subject tagging approaches for live transcription #10

New issue

Closed

opened 2026-05-13 23:59:25 +00:00 by Mark · 1 comment

Mark commented

2026-05-13 23:59:25 +00:00

Owner

Research and evaluate approaches for speaker identification and subject tagging during live recording in the transcriber app. This issue must complete before #6 can be decomposed.

Scope

Survey speaker diarization approaches: ML-based (pyannote, Cloud diarization APIs), manual tagging during recording, post-recording annotation
Evaluate subject/topic tagging: time-range based tagging, keyword detection, LLM-based topic segmentation
Research UI patterns for live tagging during recording (button press, voice activity detection)
Consider offline vs cloud tradeoffs for diarization
Look at existing Android transcription apps for UX patterns
Consider data model changes needed (speaker table, time-tag model)

Deliverable

A written recommendation: speaker identification approach, subject tagging mechanism, proposed data model changes, and UI pattern recommendation.

Blocks

transcriber#6 - No speaker/subject assignment during live recording

Post-Completion

Report findings back on transcriber#6.

Research and evaluate approaches for speaker identification and subject tagging during live recording in the transcriber app. This issue must complete before #6 can be decomposed. ## Scope - Survey speaker diarization approaches: ML-based (pyannote, Cloud diarization APIs), manual tagging during recording, post-recording annotation - Evaluate subject/topic tagging: time-range based tagging, keyword detection, LLM-based topic segmentation - Research UI patterns for live tagging during recording (button press, voice activity detection) - Consider offline vs cloud tradeoffs for diarization - Look at existing Android transcription apps for UX patterns - Consider data model changes needed (speaker table, time-tag model) ## Deliverable A written recommendation: speaker identification approach, subject tagging mechanism, proposed data model changes, and UI pattern recommendation. ## Blocks transcriber#6 - No speaker/subject assignment during live recording ## Post-Completion Report findings back on transcriber#6.