Am I right in assuming that this was audio that was picked up by a witness that was making a 911 call to report a fight?
If so, then the scenario is a teeny-tiny cell phone microphone picking up audio at a distance and out of line-of-sight (since witness did not see the fight, I have to assume it was like from a porch or through a window).
I am no audio expert, but I have sold HiFi gear all my life and the idea that distant background voices recorded by a microphone from the lowest bidder, digitized to the smallest bit-rate possible, sent through other electronic means and then laid on another recorder could have any testimony worthy information is ludicrous.
Why not?
We see it every week on the NCIS and CIS TV shows...:^)
It's possible to do some pretty amazing things with an uncompressed high-resolution audio recording or--better yet--a set of independent uncompressed high-resolution audio recordings. Detecting the correlations among small details which are too small to be heard directly may allow things which are too faint to be audible in the recordings to be reliably reconstructed. Unfortunately, audio compression techniques used in mobile phones routinely throw out lots of the smaller details without keeping any record that they've done so. Anyone who's used mobile phones much will know that they often distort people's voice in weird ways, so even having someone listen to a recording is not apt to be sufficient for certain identification.
The more fundamental question one should have with the recording is whether the screaming seems more consistent with a person whose head and phase were being battered, or with a person whose fist was being battered. One wouldn't need voice samples of the people in question to make that determination.