Personal project
Scrollkeeper
A Discord recording and transcription pipeline that turns separate speaker tracks into searchable transcripts and campaign notes.
Built the recorder and transcription workflow; extended the current application
- Python
- Node.js
- Discord.js
- Whisper
- FastAPI
- Docker
Problem
A long voice session is difficult to revisit when its useful details remain inside an audio recording.
My contribution
- Built the Discord recording and transcription workflow.
- Extended the application with transcript browsing and campaign wiki updates.
- Connected separate recording, transcription, and transcript-processing stages.
How it works
Workflow diagram
Record speakers
Capture separate Discord audio tracks for the session.
Transcribe audio
Run speech recognition over the recorded tracks.
Merge the session
Combine speaker-attributed results into a session transcript.
Browse and reuse
Search the transcript and use its content for campaign notes.
Engineering decisions
Separate pipeline stages
Recording, transcription, merging, and browsing have distinct responsibilities.
Keep speaker context
Separate tracks support transcripts that identify who said each passage.
Result and limits
The current application connects recorded sessions, searchable transcripts, and campaign wiki updates.
- Transcription and extracted notes need human review.
- The public repository contains the original recorder. The newer web application and rewrite are not published in that snapshot.