A speech recognizer that works in airplane mode

What it does

Vosk is an offline speech recognition toolkit that turns audio into text across more than 20 languages and dialects. It ships bindings for Python, Java, Node.js, C#, C++, Rust, Go, and others, and targets use cases from chatbots and smart-home devices to lecture transcription and movie subtitles. The models are advertised as 50 Mb, and the toolkit includes a streaming API, speaker identification, and a reconfigurable vocabulary.

The interesting bit

The pitch is breadth without the cloud: the same codebase is meant to run on a Raspberry Pi, an Android phone, or a server cluster. That is an unusually wide deployment target for a toolkit that also claims speaker identification and vocabulary tweaks.

Key highlights

• Offline transcription for 20+ languages and dialects, from English and Hindi to Esperanto and Kazakh
• 50 Mb models with a streaming API, speaker identification, and reconfigurable vocabulary
• Bindings for Python, Java, Node.js, C#, C++, Rust, Go, and others
• Scales from edge hardware (Android, iOS, Raspberry Pi) up to large clusters
• Use cases span chatbots, virtual assistants, smart appliances, subtitles, and interview transcription

Caveats

• The README is essentially an elevator pitch: it omits accuracy benchmarks and pushes detailed documentation to an external website.
• It is unclear from the source how reconfigurable vocabulary or speaker identification are implemented.

Verdict

Worth a look if you need speech-to-text on air-gapped devices, cheap edge hardware, or in languages cloud APIs often ignore. Give it a pass if you demand published accuracy metrics or architectural details in the repo itself.