Typr

Real-time voice transcription app with local whisper.cpp engine, live streaming transcript, and AI earpiece coaching pipeline.

Built on Tauri 2 (Rust + TypeScript) with whisper.cpp for on-device speech recognition. No cloud APIs required.

Features

• Live Streaming Transcript -- real-time text as you speak, right in the app window
• Local Whisper Engine -- whisper.cpp runs on Apple Silicon Metal GPU, fully offline
• Three Output Modes
  • 🟢 Document -- append timestamped transcriptions to daily files
  • 🟠 Clipboard -- paste transcribed text into any active app
  • 🔴 Terminal -- stream to /tmp/typr-stream for piping to other tools
• System Tray -- menu bar icon with quick controls, left-click to toggle
• Global Hotkey -- Cmd+Shift+Space to start/stop recording
• Earpiece Pipeline -- AI coaching mode: live audio → transcription → LLM → TTS response

Architecture

Microphone → whisper-stream (local, Metal GPU) → /tmp/typr-stream
                                                       ↓
                                              earpiece.sh (optional)
                                                       ↓
                                              LLM API → TTS → speaker/earpiece

Requirements

• macOS (Apple Silicon recommended)
• Rust + Node.js
• whisper.cpp (brew install whisper-cpp)

Quick Start

# Install whisper.cpp
brew install whisper-cpp

# Clone and build
git clone https://github.com/Skippy-the-Magnificent-one/typr.git
cd typr
npm install
npx tauri build

# Install the app
cp -R src-tauri/target/release/bundle/macos/Typr.app /Applications/

# Download a whisper model (first run)
# Open Typr → Engine → Download

CLI Tools

typr-live

Real-time transcription in your terminal:

typr-live              # Start transcribing (Ctrl+C to stop)
typr-live --file       # Also save to /tmp/typr-stream

typr-ctl

Remote control for Typr:

typr-ctl start         # Launch Typr
typr-ctl stop          # Quit Typr
typr-ctl listen 10     # Record 10 seconds, return transcription
typr-ctl status        # Check status
typr-ctl mode terminal # Switch output mode

earpiece

AI coaching pipeline:

earpiece                    # Start earpiece (listens + responds via TTS)
earpiece --quiet            # Text-only, no speech
earpiece --context "..."    # Custom system prompt
earpiece --buffer 3         # Process every 3 transcript lines

Configuration

Config file: ~/Library/Application Support/com.typr.app/config.json

Setting	Default	Description
engine	local	Transcription engine (local/cloud)
whisperModel	small	Whisper model size (base.en/small/medium)
recordingMode	toggle	Toggle or push-to-talk
outputMode	clipboard	Output destination (clipboard/document/terminal)
streamStep	2000	Stream processing interval (ms)
streamLength	8000	Audio window length (ms)
hotkey	CmdOrCtrl+Shift+Space	Global hotkey

Models

Model	Size	Speed	Accuracy
base.en	142 MB	Fastest	Good
small	466 MB	Fast	Better
medium	1.5 GB	Moderate	Best

Models are downloaded to ~/Library/Application Support/com.typr.app/ on first use.

Credits

Originally forked from albertshiney/typr. Extended with live streaming, output modes, system tray, earpiece pipeline, and CLI tools.

License

MIT