Uses Chrome's built-in Language Model API — speech stays on your device.
Requires a supported Chrome build with the Prompt API enabled.