That was likely Gyrophone (2014) from Stanford/Rafael researchers who demonstrated accelerometers and gyroscopes could capture speech frequencies at 80-250Hz - sufficient for speaker identification and some speech recognition without requiring microphone permissions.