Niri Transcribe

A ChatGPT-style dictation system for Linux Wayland environments, providing session-based speech-to-text transcription with immediate text input to focused windows.

🎯 Project Goals

• Session-based speech recognition with immediate text injection into active applications
• Wayland-native integration for seamless Linux desktop experience
• Docker-containerized deployment for easy setup and consistent environments
• Multiple transcription backends (OpenAI Whisper API, local models)
• User-controlled recording with start/stop session management
• Audio pipeline optimization with session buffering and format conversion

🚧 Work in Progress

This project is currently under active development. Core audio capture functionality is implemented, with transcription services and UI components in progress.

Current Status: Session-Based Dictation Ready (Tickets 001-005, 012 complete)

📖 Documentation

• Implementation Roadmap - Detailed project timeline, ticket dependencies, and progress tracking with actual vs estimated completion times
• Technology Stack - Complete technical architecture overview including Node.js audio pipeline, Docker containerization, and integration strategies
• Development Tickets - Individual implementation tickets with detailed specifications, requirements, and acceptance criteria
• QA Testing Plans - Manual testing procedures, performance benchmarks, and verification protocols for implemented features

✅ Completed Features

• Docker infrastructure with full audio stack access
• Session-based audio capture service with PipeWire/PulseAudio support
• JSON schema-based configuration system
• Audio format conversion and session buffering
• REST API endpoints for device management and recording control
• OpenAI Whisper API integration with transcription manager
• Complete session transcription workflow
• Comprehensive test suite (78 tests) and QA documentation

🔄 In Development

• Text output and injection service
• Host activation script for Niri integration
• Build and installation scripts

⚠️ Important Disclaimer

This entire solution has been "vibecoded" using Claude Code AI assistant.

While the codebase follows established patterns and includes comprehensive testing, use at your own risk. The project is experimental and may contain bugs or unexpected behavior. Always review the code before deployment in production environments.

🚀 Quick Start

# Clone the repository
git clone <repository-url>
cd niri-transcribe

# Start the service
docker compose up --build

# Check health
curl http://localhost:3000/health

📋 Requirements

• Docker and Docker Compose
• Linux with Wayland compositor
• Audio system (PipeWire/PulseAudio/ALSA)
• Optional: OpenAI API key for cloud transcription

📜 License

MIT License - see LICENSE file for details.

🤝 Contributing

This project is in early development. Feel free to open issues or submit pull requests.