Type English text, pick a voice, press Speak. The model runs in your browser; nothing is sent to a server.
The model downloads on the first Speak and is cached by the browser after that.