The radar system captures these minute movements, which are then processed by a modified version of Whisper, an AI speech-recognition model originally designed for clear audio.
Rather than retraining the entire AI system, researchers used a "low-rank adaptation" technique, tweaking just 1 percent of the model's parameters to optimize it for interpreting noisy radar data. The result? A system capable of reconstructing conversations with surprising — and unsettling — accuracy...<<<Read More>>>...
