vosk-browser

Why we included this project

Speech-to-text that runs entirely on the user's device is the main draw here: vosk-browser packages the Kaldi-based Vosk engine into a WebAssembly build you load like any other TypeScript module, so no audio ever leaves the page for a cloud API. You point it at a model file, capture the microphone, and get partial results and final transcripts back as they come. Recognition runs inside a Web Worker by design, which keeps the page responsive while the engine churns. A live demo transcribes microphone input or uploaded audio files in 13 languages, a quick way to judge accuracy with your own voice before building anything. Just bring your own Vosk model files; the library handles the plumbing around them.