Joining in late ...
If I understand Francesco right, he is pointing out that the work most likely to be distributed is the very first one mentioned above: the digitally-encoded audio. That work can be modified by editing the digital audio data in an appropriate program, which thus creates a derived work of the existing digital audio.
The GPL says "*preferred* form for modification". Part of the problem is clearly that the above is NOT the preferred form.
Re-creating the audio recording from scratch, as Rudolf describes, would be a new copyright work, and I agree that it's disingenuous to describe this as “editing”. It would be a derived work of the copyrighted words, musical arrangement, etc.; but it is less clear what its copyright relationship to the previous digital audio data would be.
THIS is the "preferred form". This is crux of this problem - there ISN'T a "preferred form for modification".
That was the point further back about a program having multiple authors yet still being a coherent work. You can't necessarily do that with art.
So what do you do here? If the original author's "preferred form of modification" is "throw it away and start again", we're in the realms of "the original author can distribute without source, and nobody else can distribute", and we're effectively saying "the GPL is useless for artists", which I think is the whole point of this discussion! :-)
Cheers, Wol -- Anthony W. Youngman - anthony@thewolery.demon.co.uk