Record meetings. Get transcripts. All on your Mac.

---

✨ Features

Feature	Description
🎙️ Live Meeting Capture	Record system audio from Zoom, Teams, Meet - transcribe locally
📁 File Transcription	Drag-and-drop audio/video files for instant transcription
🗣️ Speaker Diarization	Automatic speaker identification using local AI
🌊 Live Waveform	Visual feedback showing audio levels while recording
🎨 Dark & Light Themes	Modern floating window with customizable appearance
📜 Session History	Browse and search past transcriptions
🔒 100% Local	Audio never leaves your Mac - privacy first

🚀 Quick Start

git clone https://github.com/regmontanhani/earshot.git
cd earshot
./setup.sh

That's it. The setup script:

• Installs dependencies (BlackHole, Ollama, Python packages)
• Guides you through audio device configuration
• Pulls the local AI model for speaker identification

Then launch:

./run_window.sh

📖 How It Works

1️⃣	Configure Audio Create a Multi-Output Device in Audio MIDI Setup that routes sound to both your speakers AND BlackHole for capture.
2️⃣	Record Click Record in the floating window. The waveform shows live audio levels. Your meeting continues normally.
3️⃣	Transcribe Click Stop. Earshot transcribes using MLX-Whisper (optimized for Apple Silicon), then identifies speakers with local Ollama.
4️⃣	Review The transcript appears with speaker labels. Browse past sessions with the ◀ ▶ buttons. Open files in any format you need.

⚙️ Settings

Open settings with the ⚙️ button or Cmd+,

Setting	Options
Output Directory	Where transcripts are saved (default: ~/Documents/Earshot)
Output Formats	JSON, TXT, SRT, VTT, TSV
Model Size	tiny → turbo (larger = more accurate, slower)
Theme	Dark or Light
Opacity	50% - 100% window transparency
Always on Top	Keep window above other apps
OpenAI API Key	Optional - for cloud transcription

⌨️ Keyboard Shortcuts

Shortcut	Action
Space	Start/Stop recording
← / →	Navigate session history
Cmd+,	Open settings
Cmd+Q	Quit

📝 Output Formats

Format	Use Case
JSON	Full data with timestamps, segments, speakers
TXT	Plain text for reading or sharing
SRT	Subtitles for video editors
VTT	Web subtitles for browsers
TSV	Import into spreadsheets

🔧 Requirements

	Minimum	Recommended
macOS	12.0 Monterey	14.0+ Sonoma
Chip	Intel or Apple Silicon	Apple Silicon (M1+)
RAM	8 GB	16 GB
Disk	5 GB	10 GB

🏗️ Architecture

earshot/
├── window.py           # PySide6 floating window
├── widgets/            # UI components
│   ├── waveform.py     # Live audio visualization
│   ├── transcript.py   # Transcript viewer
│   └── settings.py     # Settings dialog
├── history.py          # Session history manager
├── themes.py           # Dark/Light stylesheets
├── audio_capture.py    # PyAudio + BlackHole
├── transcriber.py      # MLX-Whisper + OpenAI
├── diarization.py      # Ollama speaker ID
├── output_writer.py    # File output
└── config.py           # Settings

🔒 Privacy

• Audio stays local - Transcription runs on your Mac using MLX-Whisper
• Speaker ID is local - Ollama runs entirely offline
• No telemetry - No data collection, no analytics
• Optional cloud - OpenAI API only if you configure it

🤝 Contributing

Contributions welcome! See CONTRIBUTING.md for guidelines.

📄 License

MIT - see LICENSE for details.

---

Made with ❤️ for people who take too many meeting notes