vocagateway
Optional self-hosted speech-to-text headless gateway for Voca clients (beta). Provides secure, private transcription without requiring a hosted cloud service.
Ficha actualizada 23 sept 2026, 0:56 · upstream 23 sept 2026, 0:34 · release v0.1.0
Qué es vocagateway
Descripción editorial construida a partir del repositorio, su documentación y sus señales de mantenimiento.
VocaGateway is an optional self-hosted speech-to-text headless gateway designed to process audio recordings from Voca mobile and desktop applications. It uses FFmpeg for audio normalization and local speech engines such as MLX Audio, WhisperKit, and faster-whisper to transcribe audio into text. The gateway operates in a bounded environment where audio leaves the client device and travels to the machine you choose to host it on. This ensures privacy since no voice data is processed on the user's device. The WebUI provides setup assistance including model management, engine selection, microphone testing, and operational status monitoring. A CLI tool (vocagateway) allows programmatic control of the gateway through environment variables and configuration files.
- Voice dictation for visually impaired users who need real-time text transcription
- Secure private communication where voice data must never leave a trusted network
- Remote transcription services that require complete privacy and data sovereignty
- Accessibility solutions for individuals with hearing impairments or speech difficulties
- Integration with custom applications that need reliable, low-latency audio-to-text conversion
- Self-hosted architecture ensures complete control over where voice data is processed
- Apple Silicon optimized with MLX Audio and WhisperKit for fast inference on Mac
- Docker Compose provides reproducible Linux images across amd64 and arm64 architectures
- Tailscale support enables secure private network communication without exposing ports to the internet
- Comprehensive model library supporting 108 languages with multiple engines including Whisper.cpp, Moonshine, and Parakeet
- No Voca account is required but no hosted cloud transcription service is available
- Phone pairing requires a trusted network (LAN or Tailscale); public internet exposure of port 8765 is not recommended
- Mobile apps are currently unavailable; desktop embed functionality is planned but not shipped
- Performance depends on host hardware; MLX Audio and WhisperKit require Apple Silicon for optimal results
- Not all language models are available in every supported language, and model quality varies by language
Perfil de infraestructura
Estimación inicial derivada del repositorio y de los servicios detectados.
Despliegue detectado
Qué señales reales de instalación hemos encontrado.
Three deployment modes are supported: native macOS (uses MLX Audio and WhisperKit), native Linux (uses sherpa-onnx INT8 or faster-whisper), and Docker Compose for reproducible Linux images. Use `uv sync --all-groups --extra engines` to install dependencies, then run `uv run vocagateway`. For Tailscale Serve only, bind the gateway to 127.0.0.1:8765 instead of 0.0.0.0. The `/health/ready` endpoint returns 503 until a model can transcribe and 200 once ready. Model downloads are managed via the WebUI or API endpoints at /v1/models and /v1/transcriptions. For production use, consider setting up HTTPS reverse proxying (e.g., Nginx) to protect port 8765 from public exposure.
Stack y dependencias
Componentes inferidos a partir de manifests, configuración y documentación.
Puede sustituir a
Relaciones detectadas por el análisis del repositorio y su categoría.
Experiencia real de la comunidad
Ayuda a contrastar la estimación automática con instalaciones reales.
Todavía no hay datos comunitarios para este proyecto.