the machine is hearing things
A speech-recognition model is listening to a room that isn't speaking.
Whisper, running entirely inside this tab, cannot sit with silence. When the signal goes thin it invents words, whole crowds, voices that were never there. You'll watch the speech it fabricates rise off a dead-flat line and dissolve back into the noise it came from.
Nothing is recorded or sent anywhere. The model and all audio stay on your device.
This browser can't render the type field (no WebGPU). The listening still works. Open in a recent Chrome, Edge, or Safari to see it in full.