Continuous voice recording app with AI transcription and speaker detection.

Panauricon records audio in 30-minute clock-aligned segments, automatically uploads to Firebase Storage, and transcribes using Google's Gemini 2.5 Flash with intelligent speaker detection.

Want to deploy your own instance?

• Easy way: See QUICKSTART.md - Just run ./scripts/setup.sh OR ask Claude/ChatGPT to set it up
• Manual way: See SETUP.md - Detailed step-by-step guide

Takes ~5 minutes with the automated script or AI assistance.

• 30-minute segments - Auto-stops at :00 and :30, immediately restarts
• Seamless continuity - Zero-gap recording across segments and app crashes
• Clock time display - Shows current time, not elapsed duration
• Auto-resume - Restarts recording after app restarts or crashes
• Battery optimized - 24kbps AAC-LC, 16kHz, minimal notification

• Gemini 2.5 Flash - Fast, accurate, cost-effective (~$3/24hrs)
• Speaker detection - Groups speech by speaker naturally
• Clean format - Speaker paragraphs without cluttered timestamps
• Auto-retry - 3 attempts with exponential backoff for network failures

• WiFi-only uploads - Save mobile data (default ON)
• Search - Full-text search across all transcripts
• Calendar view - Visual timeline of recordings by date
• Dark mode - System-aware theme
• Local backup - Keeps last 3 audio files as safety net
• Health checks - Verifies recording every 5 minutes, auto-recovers

• Exponential backoff - Failed uploads retry at 1min → 5min → 15min → 1hr
• Battery whitelist prompt - Prevents Android from killing the app
• File existence checks - Graceful handling of missing files
• Foreground service - Reliable background recording

┌─────────────────┐
│   Flutter App   │
│  (Android)      │
└────────┬────────┘
         │
         ├─> AudioService (record)
         ├─> UploadService (WiFi check)
         ├─> RecordingManager (orchestrate)
         └─> DatabaseService (SQLite)
              │
              ▼
     ┌────────────────┐
     │ Firebase Cloud │
     │   Functions    │
     └────────┬───────┘
              │
              ├─> Cloud Storage (audio)
              ├─> Firestore (metadata)
              └─> Gemini 2.5 (transcribe)

• Flutter 3.10+
• Android SDK
• Firebase project
• Google AI API key (Gemini)

1. Clone

git clone https://github.com/eddowding/panauricon.git
cd panauricon

2. Flutter dependencies

flutter pub get

3. Firebase setup

# Add your google-services.json to android/app/
# Update lib/config.dart with your Firebase project ID and bucket

4. Deploy Cloud Functions

cd ../functions
npm install
firebase deploy --only functions

5. Run

flutter run

Recording settings (lib/config.dart):

• Segment duration: 30 minutes
• Bitrate: 24 kbps (speech-optimized)
• Sample rate: 16 kHz
• Format: AAC-LC (M4A)

Costs (Gemini 2.5 Flash):

• ~$2.50-$3.50 per 24 hours
• ~$76-$107 per month (continuous)
• WiFi-only uploads reduce mobile data costs

1. Clock boundaries (:00, :30) - seamless segment transition
2. Manual stop button - saves segment, immediately restarts
3. App crashes - detects on startup and resumes
4. Phone calls - pauses during call, resumes after

**Ed:** This is much cleaner without timestamps everywhere.
Speech flows naturally in paragraphs grouped by speaker.

**Sarah:** I agree. The old format was too verbose with
timestamps on every line.

**Ed:** Exactly. Now it's actually readable.

• Transient network failures: 3 attempts (5s, 15s, 30s delays)
• Persistent failures: Exponential backoff on upload queue
• Manual retry available for any failed transcription

lib/
├── services/        # Core business logic
│   ├── audio_service.dart
│   ├── upload_service.dart
│   ├── recording_manager.dart
│   └── api_service.dart
├── ui/
│   ├── screens/     # Main screens
│   └── widgets/     # Reusable components
├── models/          # Data models
└── config.dart      # App configuration

android/
└── app/src/main/
    ├── kotlin/      # Native Android code
    └── res/         # Resources, icons, widgets

• Frontend: Flutter 3.10, Material 3
• Backend: Firebase Cloud Functions (Node.js)
• Database: Cloud Firestore + SQLite (local)
• Storage: Firebase Cloud Storage (EU)
• AI: Google Gemini 2.5 Flash
• Location: OpenStreetMap Nominatim (reverse geocoding)

This is a personal project but open to contributions. Please open an issue before major changes.

MIT License - see LICENSE file

Built with Claude Code (Anthropic) during an intensive development session.