Vosk

Vosk is an open-source offline speech recognition toolkit designed for real-time transcription on mobile devices, desktops, and embedded systems. It provides lightweight, streaming recognition across multiple languages and runs entirely offline on CPUs. The project supports Windows, Linux, macOS, Android, iOS, and embedded platforms such as Raspberry Pi, making it suitable for client-side and server-side deployments.

The recognition engine behind Vosk is built around Kaldi-based acoustic modeling with neural network front-ends. It loads compact models that enable real-time transcription without an internet connection. Users load a language and acoustic model from the Vosk Model Registry, then feed PCM audio data to a recognizer instance. The API returns transcripts, confidence scores, and streaming results as the audio is processed.

Vosk offers bindings for several programming languages, including Python, Java, C#, and JavaScript (Node.js), enabling integration into desktop applications, mobile apps, and web services. It emphasizes offline operation, small resource footprints, and straightforward deployment. The toolkit supports multiple languages, with both official and community-contributed models available for download and use.

Licensing and availability: Vosk is released under the Apache License 2.0. It is maintained by Alpha Cephei (alphacephei) and a community of contributors, with ongoing development and updates to models and APIs. The project aims to facilitate accessible offline speech recognition for applications that require privacy, low latency, or operation without network connectivity.

See also: Kaldi-based speech recognition, offline ASR, language models, speech recognition APIs.