Transform your meetings with AI-powered real-time transcription and intelligent Q&A. Live Transcripts captures system audio, transcribes it instantly using OpenAI's GPT-4o models, and enables contextual conversations about ongoing discussions—all while your meeting continues.

Live Transcripts combines cutting-edge AI technologies to create a seamless meeting intelligence platform:

Get up and running in under 5 minutes:

# One-line setup
curl -sSL https://raw.githubusercontent.com/forayconsulting/cwmc/master/scripts/quick-setup.sh | bash

# Or manual setup
git clone https://github.com/forayconsulting/cwmc.git
cd cwmc
./scripts/configure.sh
./scripts/dev-run.sh

The setup wizard will guide you through API key configuration, audio device selection, and platform-specific requirements.

Before installation, ensure you have:

• Python 3.9+ installed
• OpenAI API Key for transcription (Get one here)
• Google AI API Key for Q&A (Get one here)
• Audio capture configured for your OS (detailed below)

macOS requires BlackHole for system audio capture. Here's the complete setup:

1. Install BlackHole

   brew install blackhole-2ch

2. Create Multi-Output Device

   • Open Audio MIDI Setup (Spotlight → "Audio MIDI Setup")
   • Click "+" → "Create Multi-Output Device"
   • Check boxes for:
     • ✅ BlackHole 2ch (required)
     • ✅ Your speakers/headphones
   • Name it "Live Transcripts"
   • Set your preferred device as Master (not BlackHole)

3. Set System Output

   • System Preferences → Sound → Output
   • Select "Live Transcripts" multi-output device
   • This routes audio to both your speakers and transcription

Windows includes built-in WASAPI loopback support—no additional setup required! The application automatically captures system audio.

Linux supports multiple audio backends with automatic detection:

# Install dependencies (choose based on your distro)
# Ubuntu/Debian:
sudo apt-get install portaudio19-dev pulseaudio

# Fedora:
sudo dnf install portaudio-devel pulseaudio

# Configure audio backend
./scripts/configure-linux.sh

Supported backends: PipeWire (recommended), PulseAudio, ALSA, JACK

Live Transcripts adapts to your specific needs through customizable Session Focus settings:

Focus: "Identify action items and decisions"

The AI will:

• Highlight assignments with owners and deadlines
• Track key decisions and their rationale
• Generate follow-up questions about implementation
• Surface risks and dependencies

Focus: "Track objections and buying signals"

Automatically:

• Identifies customer pain points and concerns
• Highlights positive buying indicators
• Suggests qualifying questions in real-time
• Tracks pricing discussions and next steps

Focus: "Capture technical details and architecture decisions"

Features:

• Accurate transcription of technical terminology
• Tracking of design decisions and trade-offs
• Integration points and dependencies
• Action items for technical implementation

Focus: "Monitor comprehension and engagement"

Provides:

• Questions that check understanding
• Identification of topics needing clarification
• Engagement metrics and participation tracking
• Summary of key learning points

# Development mode (recommended - shows all logs)
./scripts/dev-run.sh

# Or using make
make dev

# Direct Python execution
python -m src.livetranscripts.main

You'll see the startup confirmation:

✓ Audio capture initialized
✓ Batch processor initialized  
✓ GPT-4o transcription initialized
✓ Gemini integration initialized
🎤 Live Transcripts is running!
📡 WebSocket server: ws://localhost:8765
⏸️ Recording starts paused - click Start in web UI

Open http://localhost:8765 in your browser to access:

• Recording Control: Start/stop transcription
• Session Focus: Set meeting intent
• Live Transcripts: Real-time speech-to-text
• Smart Q&A: Ask questions with suggested prompts
• Insights Panel: Auto-generated summaries
• Knowledge Base: Upload reference documents

Connect via WebSocket for programmatic access:

const ws = new WebSocket('ws://localhost:8765');

// Ask a question
ws.send(JSON.stringify({
  type: 'question',
  question: 'What were the main decisions?',
  request_id: 'q123'
}));

// Receive structured response
ws.onmessage = (event) => {
  const response = JSON.parse(event.data);
  console.log(response.answer);
};

Customize behavior through environment variables:

# Transcription
TRANSCRIPTION_MODEL=gpt-4o-transcribe  # or gpt-4o-mini-transcribe, whisper-1
MODEL_FALLBACK=whisper-1                # Automatic fallback model

# Processing
MIN_BATCH_DURATION=3   # Minimum seconds before transcription
MAX_BATCH_DURATION=30  # Maximum batch size
SILENCE_THRESHOLD=500  # Milliseconds of silence to trigger batch

# AI Models
GEMINI_MODEL=gemini-2.0-flash-lite  # Optimized for rate limits
INSIGHT_INTERVAL=60                 # Seconds between auto-insights

# Server
SERVER_HOST=localhost
SERVER_PORT=8765

Your data remains secure through multiple layers of protection:

• Local Processing: Audio analysis happens on your device
• No Storage: Audio is never saved to disk
• API-Only: Only transcribed text sent to cloud services
• Encrypted Keys: API credentials stored locally in .env
• Session Isolation: Each session's context is independent

Typical usage costs per hour:

Note: Costs scale with speaking time, not meeting duration

No Audio Capture

• macOS: Verify "Live Transcripts" is selected as system output
• Windows: Check default audio device in Sound settings
• Linux: Run ./scripts/list-audio-devices.sh to verify device

Poor Transcription Quality

• Ensure clear audio without echo/feedback
• Check microphone levels aren't clipping
• Minimize background noise
• Verify correct audio device selected

API Errors

• Confirm API keys in .env file
• Check rate limits in provider dashboards
• Verify internet connectivity
• Review error logs for specifics

For large meetings or multiple speakers:

# Increase batch size for stability
MAX_BATCH_DURATION=45

# Reduce insight frequency  
INSIGHT_INTERVAL=120

# Adjust silence detection
SILENCE_THRESHOLD=750

┌─────────────────┐     ┌──────────────────┐     ┌─────────────────┐
│  Audio Capture  │────▶│  VAD Batching    │────▶│ GPT-4o/Whisper  │
│  (Platform API) │     │  (Smart Chunks)  │     │ (Transcription) │
└─────────────────┘     └──────────────────┘     └─────────────────┘
                                                           │
                                                           ▼
┌─────────────────┐     ┌──────────────────┐     ┌─────────────────┐
│   Web UI/API    │◀────│  Gemini Q&A      │◀────│ Context Manager │
│  (WebSocket)    │     │  (Insights)      │     │  (Full History) │
└─────────────────┘     └──────────────────┘     └─────────────────┘

We welcome contributions! Key areas for enhancement:

• Additional audio backend support
• Enhanced language support
• Custom insight generators
• Integration plugins
• UI improvements

See CONTRIBUTING.md for guidelines.

This project is licensed under the MIT License. See LICENSE for details.

Built with ❤️ by teams who believe every conversation holds valuable insights.