Back to Gap Map Model components / Speech & audio

Vosk

open core / Overall score: 3.0

Alpha Cephei's offline speech recognition toolkit built on Kaldi: a streaming recognizer for 20+ languages with small models for phones and Raspberry Pi boards and large ones for servers, speaker identification, and bindings for Python, Java, Node.js, C#, Go, Rust and more.

The last PyPI release is from December 2022 although the repository is still updated; most downloadable models are Apache-2.0, and a few carry AGPL, LGPL or non-commercial terms.

Openness

4 low confidence

4.0

license: Apache-2.0(OSI
source: public
core features withheld: yes — Alpha Cephei sells Vosk Enterprise and Vosk Mobile tiers with more languages, premium-accuracy models and iOS and Unity support that the open-source edition does not carry

The Vosk engine is Apache-2.0 and builds from the published repository. Alpha Cephei also sells Enterprise and Mobile editions that add languages, higher-accuracy models and platforms the open-source release lacks, so the open code is not the whole product.

https://alphacephei.com/en/ recorded 2026-09-27
Alpha Cephei products: "Vosk Enterprise more languages premium accuracy ongoing support technology training"; "Vosk Mobile more languages efficiency for the edge iOS and Unity".
https://raw.githubusercontent.com/alphacep/vosk-api/master/COPYING recorded 2026-09-27
COPYING body: "Apache License" / "Version 2.0, January 2004".
https://raw.githubusercontent.com/alphacep/vosk-api/master/README.md recorded 2026-09-27
README: "Vosk scales from small devices like Raspberry Pi or Android smartphone to big clusters."

Adoption

3 high confidence

3.0

PyPI downloads of the vosk package; the Android, Java and Node bindings install through other registries not counted here.

https://pypistats.org/api/packages/vosk/recent recorded 2026-09-27
"last_month":572791 for vosk.

Capability

3 medium confidence

3.0

A dependable offline recognizer that runs almost anywhere, but it serves one task on Kaldi-era models, two steps below NeMo.

https://alphacephei.com/vosk/models recorded 2026-09-27
Models page: "You can check Vosk recipe for details: https://github.com/alphacep/vosk-api/tree/master/training".
https://raw.githubusercontent.com/alphacep/vosk-api/master/README.md recorded 2026-09-27
README: "It enables speech recognition for 20+ languages and dialects"; "zero-latency response with streaming API, reconfigurable vocabulary and speaker identification."

Verified 2026-09-27