MeetMemo

AI-powered meeting transcription with speaker diarization and intelligent summarization

Features • Quick Start • Usage • Documentation

---

A meeting transcription application that runs entirely offline. It converts speech to text, identifies different speakers (diarization), and generates intelligent summaries of discussions. Connect it to your local LLM server for customized summarization. Perfect for meetings, interviews, lectures, or any audio where you need a clear transcript and actionable insights.

Features

Feature	Description
Audio Recording & Upload	Record meetings directly in browser or upload files (MP3, WAV, M4A, FLAC, WebM, OGG)
Speech Recognition	faster-whisper with CTranslate2 (4x speedup, 99+ languages)
Speaker Diarization	PyAnnote.audio 3.1 for automatic speaker identification and labeling
Audio Playback & Sync	Built-in audio player with transcript synchronization - click any segment to jump to that timestamp
AI Summarization	LLM-powered summaries with key points, action items, and insights
Real-time Progress	Live status updates and job management for long-running tasks
Speaker Management	Edit speaker names with persistent storage across sessions
Export Options	Professional PDF and Markdown exports for transcripts and summaries
Multi-language	Automatic language detection or specify target language

Quick Start

Prerequisites

Software	Version	Purpose
Docker	Latest	Container runtime
Docker Compose	Latest	Multi-container orchestration
NVIDIA GPU	CUDA 12.1+	ML model inference (optional, CPU fallback available)
LLM Server	Any OpenAI-compatible API	Summarization (e.g., LM Studio, Ollama, OpenAI)

Minimum Hardware:

• GPU: 4GB VRAM (8GB+ recommended)
• RAM: 8GB (16GB+ recommended)
• Storage: 10GB (for models and data)

Installation

1. Clone and setup:

git clone https://github.com/NotYuSheng/MeetMemo.git
cd MeetMemo
cp example.env .env

2. Accept Hugging Face model licenses:

• Visit speaker-diarization-3.1 → Accept
• Visit segmentation-3.0 → Accept
• Create token at HF Tokens with Read access

3. Configure .env:

# Required
HF_TOKEN=hf_your_token_here
LLM_API_URL=http://localhost:1234
LLM_MODEL_NAME=qwen2.5-14b-instruct
LLM_API_KEY=

# Optional
POSTGRES_PASSWORD=changeme      # Change in production!
TIMEZONE_OFFSET=+8              # Your timezone

4. Start the application:

docker compose up -d

5. Access MeetMemo:

Open https://localhost in your browser.

Note: You'll see a certificate warning (self-signed SSL). Click "Advanced" → "Proceed" - this is expected and required for microphone access.

Usage

Basic Workflow

graph LR
    A[Upload/Record Audio] --> B[Transcribe]
    B --> C[Review Transcript]
    C --> D[Edit Speaker Names]
    D --> E[Generate Summary]
    E --> F[Export PDF/Markdown]

Loading

1. Upload or Record - Upload an audio file or record directly in browser (HTTPS required)
2. Transcribe - Click "Start Transcription" to process with Whisper + PyAnnote
3. Review - View diarized transcript with speaker labels and timestamps
4. Playback - Use the audio player to listen while following along with highlighted transcript segments
5. Customize - Click speaker names to rename them (persists across sessions)
6. Summarize - Generate AI summary with key insights and action items
7. Export - Download professional PDF or Markdown files

Supported Audio Formats

MP3, WAV, M4A, FLAC, WebM, OGG (max 100MB default)

Tech Stack

Component	Technology
Backend	FastAPI, Python 3.10+, Uvicorn, Pydantic Settings
Architecture	Layered architecture (API → Service → Repository → Database)
Frontend	React 19, Vite, Lucide Icons
Reverse Proxy	Nginx with SSL/TLS (self-signed certs included)
ML Models	faster-whisper with CTranslate2, PyAnnote.audio 3.1
Database	PostgreSQL 16 with asyncpg
Containerization	Docker, Docker Compose, NVIDIA Container Toolkit
PDF Generation	ReportLab, svglib

Documentation

Comprehensive documentation is available in the docs/ directory:

Document	Description
Architecture	System architecture, design patterns, data flow
API Reference	Complete REST API documentation with examples
Configuration	Environment variables, model selection, settings
Database	Schema, queries, backup/restore, maintenance
Deployment	Production deployment with HTTPS options
Development	Developer guide for contributing
Troubleshooting	Common issues and solutions

Common Tasks

View Logs

# All services
docker compose logs -f

# Specific service
docker compose logs -f meetmemo-backend

Restart Services

# Restart all
docker compose restart

# Restart backend only
docker compose restart meetmemo-backend

Backup Data

# Backup database
docker exec meetmemo-postgres pg_dump -U meetmemo meetmemo > backup.sql

# Backup all volumes
sudo tar -czf meetmemo_backup.tar.gz /var/lib/docker/volumes/meetmemo_*

Access Database

docker exec -it meetmemo-postgres psql -U meetmemo meetmemo

Deployment

MeetMemo includes HTTPS with self-signed certificates out of the box. For production:

• Internal/Development: Use built-in self-signed certs (works immediately)
• Production: Replace with real certificates or use Cloudflare Tunnel

See Deployment Guide for detailed instructions.

Security

• Local Processing: Transcription runs entirely on your server
• Data Privacy: Audio never leaves your infrastructure (except LLM summarization)
• HTTPS: SSL/TLS enabled by default
• Database: PostgreSQL not exposed outside Docker network
• No Authentication: Add auth layer for multi-user deployments

Performance

• GPU Acceleration: Automatic CUDA support for faster processing
• Model Caching: ML models loaded once at startup
• Async I/O: All operations use async/await for concurrency
• Background Cleanup: Automatic cleanup of old jobs and exports

Troubleshooting

Having issues? Check the Troubleshooting Guide for solutions to:

• Microphone/recording not working
• GPU not detected
• Model download failures
• Container startup issues
• Performance problems

Contributing

Contributions are welcome! Please see Development Guide for:

• Development setup
• Code structure
• Testing guidelines
• Commit conventions

Sample Outputs

The sample-files/ directory contains example outputs:

• Sample audio files (MP3, WAV)
• Generated transcripts (PDF, Markdown)
• AI summaries with action items
• Application demo GIFs

Star History

If you find MeetMemo useful, consider giving it a star! ⭐

License

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