Overview

• A speech recognizer built on Vosk that can be run on the browser, inspired by vosk-browser, but built from scratch and no code taken!
• Designed with basic/nothrow exception safety
• See the usage folder's README.md for API documentation, notes.
• See the devel folder for the newest build (not guaranteed to work) and the JS build script

Additions to vosk-browser:

• Support multiple models
• Model storage path management
• Model ID management (for model updates)
• Smaller JS size (>3.1MB vs 1.4MB)
• All related files (pthread worker, worklet processor,...) are merged
• Shorter from-scratch build time

Disclaimer:

• Will log empty statement into JS console, see global.cc line 34
• Model loading can be slow due due to slow OPFS reading speed

Basic usage

• Use the prebuilt binary through JsDelivr