Firstly there are clever techniques to improve the effective resolution of the PWM
from say 8 bits or less to 16 or more by "noise shaping". This allows a high PWM frequency
to be used that can be filter out from the audio. The 16MHz frequency of the Arduino
allows 8 bit PWM to run at 62.5kHz, which is ultrasonic and thus easy to filter out
(the speaker and your ears will do the job!)