IAC-099 Module 1: Voice Recorder

Privacy-first voice recording and transcription for crisis support.

Purpose

Records voice locally and transcribes using Whisper.cpp with Metal GPU acceleration. No cloud dependencies. Works offline. Works when it matters.

Input/Output Contract

Input: User's voice (microphone) Output:

• raw-audio.wav - Original audio recording
• transcript.md - Transcribed text in markdown format

Tech Stack

• Framework: Tauri 2
• Frontend: Svelte 5
• Transcription: Whisper.cpp (C++ core, Metal GPU)
• Database: SQLite (local storage)
• Platform: macOS (Apple Silicon M2)

Prerequisites

• macOS with Apple Silicon (M2/M3)
• Node.js 20 (via nvm)
• Rust (latest stable)
• Python 3.11+ with uv (for any Python utilities)

Installation

# Use Node 20
nvm use 20

# Install dependencies
npm install

# Set up database
npm run db:migrate

Development

# Start dev server
npm run tauri:dev

# Run tests
npm test

# Lint code
npm run lint

Building

# Build production app
npm run tauri:build

Usage

1. Click "Record" button
2. Speak into microphone
3. Click "Stop" when done
4. App transcribes using Whisper.cpp (local, GPU-accelerated)
5. Outputs saved to configured directory:
   • raw-audio.wav
   • transcript.md

Configuration

• Recording settings (sample rate, format)
• Whisper model selection (tiny, base, small, medium)
• Output directory path
• Audio quality settings

Source Code References

Based on working implementation from:

• /Users/kjd/01-projects/IAC-031-clear-voice-app/

Success Criteria

• ✅ Records voice
• ✅ Transcribes to markdown
• ✅ Saves locally
• ✅ Works offline
• ✅ No cloud dependencies

Architecture

src/
├── components/      # Svelte UI components
├── stores/          # Svelte stores for state
├── lib/            # Shared utilities
└── tauri/          # Rust backend
    ├── audio/      # Recording logic
    ├── whisper/    # Whisper.cpp integration
    └── storage/    # SQLite operations

Privacy & Security

• All processing happens locally
• No data leaves the device
• No cloud API calls
• No telemetry or analytics
• User data stored only on local filesystem

License

[To be determined]

Contributing

See main project repository for contribution guidelines.