Echoloop

EchoLoop — Real-Time Meeting Copilot

Live meeting transcription + AI coaching. Captures dual audio (system speakers + microphone), transcribes in real-time, and feeds rolling transcript to Claude or GPT for tactical insights every ~35 seconds.

When to Use This Skill

• Live meeting coaching: Real-time tactical advice during calls
• Meeting transcription: Dual-stream (you + them) labelled transcript
• Meeting recap: Session logs with markdown export
• Negotiation support: Reads the room and suggests responses
• Multi-user audio: Works with our VNC desktop + virtual audio cables

When Not to Use

• For pre-recorded audio/video processing — use ffmpeg-processing
• For generating podcasts from text — use notebooklm (audio overviews)
• For video production — use open-montage
• For browser-based meeting automation — use playwright

Architecture

┌───────────────┐     ┌──────────────┐     ┌──────────────┐     ┌────────────┐
│ Audio Capture  │────>│  Transcriber │────>│   EchoLoop   │────>│     UI     │
│ (2 threads)    │     │ (thread pool)│     │   Engine     │     │  (Tkinter) │
│                │     │              │     │  (asyncio)   │     │            │
│ system ──┐     │     │faster-whisper│     │              │     │ Always-on- │
│ mic ─────┤     │     │  or Deepgram │     │ Claude / GPT │     │ top overlay│
│          v     │     │              │     │ every ~35s   │     │            │
│  chunk queue   │     │  seg queue   │     │ + on silence │     │ insight log│
└───────────────┘     └──────────────┘     └──────────────┘     └────────────┘

Quick Start

# Install dependencies
pip install -r requirements.txt   # from the echoloop skill directory

# Set API key (uses existing container keys)
export ANTHROPIC_API_KEY="$ANTHROPIC_API_KEY"  # Already set for devuser

# Configure audio devices (container VNC)
export ECHOLOOP_SYSTEM_DEVICE="pulse"     # PulseAudio virtual cable
export ECHOLOOP_MIC_DEVICE=""             # Default mic

# Launch on VNC Display :1
DISPLAY=:1 python3 -m echoloop

Container Audio Routing

Our container has PulseAudio on VNC Display :1. For meeting capture:

1. Virtual audio cable: Route system audio through PulseAudio monitor
2. Set device: ECHOLOOP_SYSTEM_DEVICE=<pulse monitor name>
3. Launch: DISPLAY=:1 python3 main.py # from the echoloop skill directory

For multi-user scenarios, each user can run their own EchoLoop instance with different audio routing.

Environment Variables

Features

• Dual audio capture: System audio + microphone, labelled [THEM] and [ME]
• Transcription: Local faster-whisper (CPU/CUDA) or Deepgram cloud
• LLM coaching: Claude or GPT every ~35s with nudge button for instant advice
• Silence detection: Skips dead air, saves CPU and API calls
• Session logging: Transcript + insights saved to disk, markdown export
• Auto-reconnect: Audio streams recover if device disconnects
• Pause/Resume: Click or Ctrl+Shift+E global hotkey
• Live stats: Insight count, session duration, talk-time percentage

Integration with Other Skills

Attribution

EchoLoop by Pete Woodbridge. MIT License.