d-scribe

Record Discord voice channel audio and transcribe it with speaker attribution using Whisper.

Features

• Capture audio from Discord (loopback + microphone) via WASAPI on Windows
• Track speakers using Discord RPC speaking events
• Transcribe segments with whisper.cpp
• Export to SRT or VTT
• Auto-save sessions to a recent folder (configurable retention, default 10 days)
• Playback with remote/local/both modes, auto-scroll transcript during playback
• Project list from default location; click to open, delete with optional audio cleanup
• Discord auth persisted via refresh token; auto-reconnect on startup

Prerequisites

• Node.js & npm
• Rust (for Tauri)
• Discord running (for RPC connection)
• Whisper binary – required for transcription (see below)

Quick Start

1. Install dependencies

npm install

2. Set up Whisper (required for transcription)

Windows (recommended):

.\src-tauri\binaries\download-whisper.ps1

Or see src-tauri/binaries/README.md for manual setup.

3. Download a Whisper model

Run the app, open Settings, and download a model (e.g. ggml-base.en.bin). Models are stored in %APPDATA%/d-scribe/models/.

4. Run the app

npm run tauri dev

5. Connect & record

1. Open Settings → connect to Discord (Client ID, Client Secret, RPC Origin). Auth is saved; you typically only need to authorize once.
2. Join a voice channel
3. Click Start Recording
4. When done, click Stop Recording – the session is auto-saved to recent
5. Use Play to listen; transcript auto-scrolls with playback
6. Click Transcribe to run Whisper on each segment
7. Save Project to move to permanent storage, or Export to SRT/VTT

Project Structure

d-scribe/
├── src/                 # React frontend (Vite + TypeScript)
├── src-tauri/
│   ├── src/             # Rust backend
│   │   ├── lib.rs       # Tauri commands, transcription orchestration
│   │   ├── audio/       # WASAPI capture
│   │   ├── discord_rpc/ # Discord RPC, OAuth, token persistence
│   │   ├── session/     # Recording, segments, merge buffer
│   │   ├── project.rs   # Save/load, auto-save, purge, delete
│   │   └── transcription/ # WAV extraction, Whisper CLI
│   └── binaries/        # whisper-cli.exe + DLLs (run download script)
└── docs/

Data Locations

• Projects: %APPDATA%/d-scribe/projects/ (permanent saves)
• Recent sessions: %APPDATA%/d-scribe/projects/recent/ (auto-saved; purged by retention)
• Models: %APPDATA%/d-scribe/models/
• Transcription temp: %APPDATA%/d-scribe/transcribe_temp/

Settings

• Project name template: Placeholders {guild}, {channel}, {timestamp}, {date}, {time} for session IDs and filenames
• Recent sessions retention (days): How long auto-saved sessions are kept (default 10)
• Segment merge buffer (ms): Min silence before splitting segments
• Playback mode: Remote, Local, or Both (default Both; persisted)

Build

npm run tauri build

Troubleshooting

Enable debug logging (to diagnose Discord RPC, transcription, etc.):

• PowerShell: $env:RUST_LOG="d_scribe=debug,wasapi=warn"; npm run tauri dev
• Cmd: set RUST_LOG=d_scribe=debug,wasapi=warn && npm run tauri dev

(Plain RUST_LOG=debug floods the terminal with WASAPI trace logs.)

Zero segments after recording: Segmentation comes from Discord RPC speaking events. Ensure you're connected in Settings and in the voice channel before recording. If you get 0 segments, the RPC subscription or connection may need debugging.

Planned Features

• AI summary – Generate summaries of transcripts using AI/LLM
• Participant stats – Word counts and speaking time per participant
• Meeting notes workflow – Supported flow for creating structured meeting notes from transcripts
• Debate/discussion analysis – Workflow for analyzing debates and discussions (arguments, positions, rebuttals)
• Custom workflows – User-defined workflows with configurable LLM instructions (e.g. custom prompts, output formats)
• Flexible AI/LLM support – Use both local models (e.g. Ollama, llama.cpp) and remote APIs (OpenAI, Anthropic, etc.) for summarization and other tasks
• Privacy: mute-aware recording – Listen for Discord mute events; when the current user or others mute, disable the corresponding local audio stream so muted participants are not recorded

License

MIT