EchoBridge
A local-first meeting recording and transcription platform with speaker identification via voiceprints, designed for complete data privacy and control.
Why?
I started EchoBridge as a side project to solve my own need for tracking recurring meetings while maintaining complete control over sensitive data. The focus was on building a truly local-first solution that could run fine-tuned models directly on device, eliminating privacy concerns while still delivering powerful features.
Demo
EchoBridge Calendar Integration Demo
Preview
EchoBridge mobile app and transcript view
Tech Stack
Key Features
- Google Calendar Integration: Select events for automatic recording with speaker identification via voiceprints - perfect for tracking recurring meeting history
- On-premise/On-device Deployment: For companies wanting data privacy, deploy via Docker and connect to your own server with customizable models for transcription/diarization/embeddings
- Prompt Builder: Customize prompts and context (e.g. “summarize progress and next steps from last 3 weekly meetings”)
- Team Sharing: Securely share meeting recordings and transcripts with team members
- Offline-First Architecture: Complete functionality without an internet connection, with custom-built local-first engine for optimal performance
- Speaker Identification: Identifies speakers across meetings using voiceprints, building a history of participation
Tech Implementation
- Custom Local-First Engine: Built from scratch for optimized performance without external libraries
- React Native & Expo: Cross-platform mobile and web application framework
- TypeScript: For type-safe code and better developer experience
- On-device ML Models: Fine-tuned models that run directly on the device for privacy and performance
- GPU Acceleration: Optimized for GPU-intensive tasks when available
- Docker: For optional on-premise deployment
Tech Challenges
- Fine-tuning ML Models: Migrating from server-based models (SpeechBrain, Pyannote) to optimized on-device versions
- GPU-intensive Task Scaling: Optimizing performance for computationally intensive audio processing tasks
- Custom Local-First Engine: Building a complete offline-first solution from scratch for maximum optimization
- Cross-platform Consistency: Ensuring the app works seamlessly on web, Android, and iOS
- Voiceprint Recognition: Developing reliable speaker identification that works across multiple meetings
Open Source Status
The recording library is already public at GitHub with growing adoption. The backend is shared selectively since it contains core IP. The key focus is giving users full data control by running models locally (with GPU recommended for performance).