transcript editor

[ComputelessComputer]

Overview

A comprehensive transcript editor that enables users to review, edit, and refine transcriptions with speaker identification, voice prints, and memory-based learning. Inspired by Descript's editing experience.

Core Features

Speaker Identification & Segmentation

• Automatic speaker segmentation during transcription
• Smart speaker tagging that handles over-segmentation (e.g., if 2 people are wrongly segmented into 4, provide easy way to merge)
• Voice print support for recognizing recurring speakers across sessions
• Related: speaker diarization #956 (speaker diarization), voiceprint #2266 (voiceprint)

Descript-like Editing Experience

Speaker Labels

• Click on speaker label to hear a 10-second audio snippet (if recording is saved)
• If no recording available, speaker is still selectable but no audio preview
• Pop-over with audio playback controls when clicking speaker

Transcript Editing

• Double-click on transcript content to enter edit mode
• Inline editing of transcribed text
• Easy correction of mistranscribed words
• Cmd+F / search and replace functionality

Memory & Learning System

• Track user corrections to mistranscribed words
• When user changes "Hyperno" (H-Y-P-E-R-N-O) to "Hyprnote" (H-Y-P-R-N-O-T-E), remember this correction
• Two implementation approaches:
  1. Keyword boosting: Boost corrected words in future transcriptions
  2. Rule-based system: Apply correction rules automatically
• Related: memory #3474 (memory), implement keyword boosting #3512 (implement keyword boosting)

This creates an experience where "Hyprnote evolves together with you" - learning your vocabulary, jargon, and preferences over time.

Technical Context

• Transcript data stored as Word2 objects with timestamps
• Integration with ListenerActor for real-time updates
• Audio playback requires coordination with stored recordings
• Memory system needs persistent storage for corrections

Acceptance Criteria

Speaker Identification

• Automatic speaker segmentation works during transcription
• Users can merge incorrectly segmented speakers
• Voice prints can be saved and recognized across sessions
• Speaker labels are clearly displayed

Editing Experience

• Click speaker label → 10-second audio snippet plays (if recording exists)
• Double-click transcript → inline edit mode
• Cmd+F opens search with replace functionality
• Changes save automatically

Memory System

• Corrections are tracked and stored
• Corrected words are boosted in future transcriptions
• Rule-based corrections apply automatically
• User can view/manage their correction history

Related Issues

• speaker diarization #956: speaker diarization
• voiceprint #2266: voiceprint
• memory #3474: memory
• implement keyword boosting #3512: implement keyword boosting
• Make (double)-clicking on transcript switch to edit mode #1430: Make (double)-clicking on transcript switch to edit mode

Labels

product/desktop, area/ui, area/backend