4 ms·

I guess you need a good model of their head to post process a readable message out of the distorted audio.