lichess-org/vosk-browser

Vosk-Browser lets you add speech recognition directly to a website, entirely in the user's browser. Instead of sending audio to a cloud service like Google or Amazon for transcription, the audio is processed locally on the user's own device. This means you can build features like voice commands, live captioning, or voice-driven search without worrying about server costs or latency, and it works even without an internet connection once the page is loaded. The library works by running Vosk, an open-source speech recognition engine, compiled into WebAssembly, a format that lets code written in languages like C++ run inside a web browser at near-native speed. It specifically runs inside a Web Worker, which is essentially a background thread that keeps the speech recognition from freezing up the rest of your webpage. You load a language model, connect the library to the microphone or an audio file, and it starts sending back text results in real time. This is useful for web developers building apps that need voice input without the complexity or privacy concerns of a third-party cloud API. For example, someone building a language-learning app could use it for pronunciation feedback. A productivity tool could add voice commands. A video platform could offer live captions. The live demo shows it working in 13 languages, transcribing both microphone input and uploaded audio files directly in the browser. One notable tradeoff: running speech recognition entirely on the client means the user's device does all the heavy lifting, so performance depends on their hardware. The library is also still early in its lifecycle, the to-do list includes basics like automated tests and better documentation, so it may require some patience to get up and running.