Vosk
An offline speech recognition toolkit under Apache-2.0 with streaming, models of about 50 MB, 20+ languages, a reconfigurable vocabulary and speaker identification, from small boards to clusters.
What it does
Vosk "is an offline open source speech recognition toolkit". Its README lists more than 20 languages and dialects, from English and Indian English through German, French, Spanish, Portuguese, Chinese, Russian, Turkish, Vietnamese, Italian, Dutch, Catalan, Arabic, Greek, Farsi, Filipino, Ukrainian, Kazakh, Swedish, Japanese, Esperanto, Hindi, Czech and Polish.
The design point is size and latency: "Vosk models are small (50 Mb) but provide continuous large vocabulary transcription, zero-latency response with streaming API, reconfigurable vocabulary and speaker identification." It scales "from small devices like Raspberry Pi or Android smartphone to big clusters".
What it replaces
Vosk replaces Deepgram and AssemblyAI for live recognition on constrained hardware, where a 50 MB model and a streaming API matter more than the last point of accuracy.
Where it falls short: the README is a pointer to the website, so word timestamps, speaker labelling beyond identification, quantisation, and the licence of each model are not stated here. The latest tagged release is from April 2024, though the repository had commits in August 2026.
How it runs
Through language bindings: Python, Java, Node.js, C#, C++, Rust, Go and others. Installation and examples are on the website. No Docker image, server or GPU requirement is described in the README.
Who it fits
A team putting recognition on a device, a kiosk or a board, or a chatbot that needs to react as the person speaks. A team wanting the highest accuracy on recordings should use Whisper or Parakeet.
Limits the README states
None listed.
Sources
- Vosk README and Apache-2.0 LICENSE: github.com/alphacep/vosk-api, read 2026-09-16.
More speech-to-text engines
OpenAI's general-purpose speech recognition model, released with code and weights under MIT, in six sizes from 39 million to 1.55 billion parameters, with multilingual transcription, translation to English and language detection.
A dependency-free C and C++ port of Whisper under MIT that runs on CPUs, Apple Silicon and phones, with quantised models, an HTTP server and bindings for a dozen languages.
Batched Whisper transcription under BSD-2-Clause with word-level timestamps from forced alignment and speaker labels from pyannote, the closest open-source match to a hosted transcription API's output.
NVIDIA's speech models and toolkit: Apache-2.0 code, and the Parakeet TDT 0.6B v3 model under CC-BY-4.0 with 25 European languages, automatic language detection, punctuation and word timestamps.
A runtime under Apache-2.0 from the next-generation Kaldi team that runs speech to text, text to speech, speaker diarization and voice activity detection locally on CPUs, NPUs and phones, with no internet connection.
An on-device speech toolkit under MIT built for live streaming, with speech-to-text models trained from scratch in sizes down to 1 MB, one library across Python, the browser, phones and desktops.
Added September 16, 2026. Found an error on this page? Write to reveneau@licheo.com and it gets fixed in the next weekly pass.
Running one of these inside your own environment, with your own data and your own security rules, is the kind of work Reveneau does. Read how a forward deployed engagement works.