3 ms·

I believe they mean it hears human speech by applying beamforming on the sound from multiple microphones.