[userbinator]

Music is basically never going to sound good over a POTS because it's a low-bandwidth channel optimised for human voice:

http://en.wikipedia.org/wiki/Voiceband

It may also be a legacy system using old worn-out looping casette tapes for the music.

[yourcelf]

It's even worse if the sound is going through a digital (VOIP / cell phone) system. Most modern codecs for voice are using some form of Linear Predictive Coding[1] (e.g. ACELPC) which is basically modeling sound as a resonator at the bottom of a tube with a filter bank (sortof like your voice box). With voice, this is a reasonably good approximation, and the codecs are aggressively tuned to be efficient at that. But if full-band music gets piped through it will sound roughly like that music is being produced by a flapping plosive at the bottom of a long tube.

[1] https://en.wikipedia.org/wiki/Linear_predictive_coding

[spdustin]

"Flapping plosive at the bottom of a long tube."

What an exceptionally colorful description of the effects of that type of codec! Well done.

I wonder if there are applications for psychoacoustic modeling approaches for safety/compulsory announcements. Can they be encoded in a different way before broadcast on the PA to take advantage of our perception of human speech? And possible a different codec still for users of headphones?

Or perhaps a quick, sharp attention tone before urgent broadcasts, but for non safety/compulsory announcements, a lower volume setting...