hush is a browser extension that lets you easily convert speech to text, powered by Whisper and WebGPU.
I couldn't find a simple, easy to use speech to text extension for the browser that leverages WebGPU instead of querying some cloud based model, so I built one in an hour or so.
The main features are recording speech into text from the popup view, and spinning up a recording session from any text field on the page via a hotkey.
git clone https://github.com/terror/hush
bun install
bun run build
The development build should be present in .output relative to the project
root.
n.b. You'll need to allow audio recording permissions to the extension before use.
whisper-web - ML-powered speech recognition directly in your browser.