Conservation of identity when using lossy methods works differently with sound than with images.
Will I recognize this sound if I put it through image changes?
The sound’s intelligibility survived “convert to binary” (2 bit) and doing an outline of the binary forms.
but it did not survive a number of other ways. Erosion was the most fascinating though, as it played every sound the test voice I used was capable of all at once but in the rhythm of speaking.
===