Vosk

Vosk makes speech recognition work where the cloud cannot reach. The toolkit from Alpha Cephei performs fully offline transcription in more than 20 languages and dialects, from English and Spanish to Chinese, Russian, and Japanese, using compact models around 50 MB that run on hardware as small as a Raspberry Pi or an Android phone while also scaling up to server clusters. Its streaming API returns partial results with effectively zero latency, and it supports large-vocabulary continuous transcription, runtime-reconfigurable vocabularies for command grammars, and speaker identification. Bindings cover Python, Java, Node.js, C#, C++, Rust, Go, and more, which is why it shows up in everything from smart home devices and IVR systems to subtitle generators and lecture transcription tools. Released under the Apache 2.0 license with freely downloadable models, Vosk is a frequent choice for developers who need private, on-device voice input without per-minute API fees, accepting somewhat lower accuracy than large cloud models in exchange.