finds.dev

alphacep/vosk-server

WebSocket, gRPC and WebRTC speech recognition server based on Vosk and Kaldi libraries

View on GitHub

★ 1,259PythonApache-2.0updated Jul 2025

// the take

A self-hosted speech recognition server built on Kaldi/Vosk that speaks WebSocket, gRPC, WebRTC, and MQTT. Aimed at developers who need offline ASR for telephony (Asterisk, FreeSWITCH), smart home systems, or browser-based transcription without sending audio to a cloud service. The model runs locally; you own the data.

Covers every realistic transport protocol you'd need for streaming audio — the WebSocket server is the obvious entry point but the gRPC variant is what you'd actually use in production with multiple concurrent connections. Docker images are pre-built per language (en, ru, de, fr, cn, hi, ja, es) so you can skip the painful Kaldi build process entirely. Client samples span Python, Go, Java, C#, JavaScript, and PHP, which means you can drop it in front of nearly any existing stack. GPU batch mode is included for high-throughput scenarios, which most open-source ASR servers leave as an exercise.

The documentation is almost entirely offloaded to an external website — the README is two paragraphs pointing elsewhere, so you're one broken link away from flying blind. The Angular and Vue client demos look abandoned (Angular 8 era, Vuetify 2), which signals the browser integration path hasn't been seriously maintained. No built-in auth or token validation on any of the server variants; if you expose a WebSocket endpoint you're on your own for access control. The multi-language Docker images are individually maintained Dockerfiles with no shared base layer — adding a new language or updating the Vosk version means touching fifteen files.

We read the README, the directory tree and a key source file in July 2026. Nobody pays to be listed. How we judge →

// at a glance

Stars
1,259
Forks
321
Language
Python
License
Apache-2.0
Last commit
25 Jul 2025

// nearby

If vosk-server isn't quite it.

All repos →