Speak and live captions appear on a floating card. Speech recognition runs on this device with Moonshine.
Use Chrome with a microphone. Download the ~94 MB model once, then load it from the browser cache.
Captions can also be translated into Spanish, French, German or Mandarin with Opus-MT, on this device too. Each language is a separate download.
Audio stays on this device. No cloud, no API key.
Preparing the scene.
Model ready. Choose Continue in simulator, or ENTER XR when available.