Months ago I've listened to Linear Phase vs Minimum Phase filters on my SoundblasterX G6 via JDS Atom headphone amp and AKG K702 headphones, and I was quite sure at the time that I liked the option "Fast Roll Off - Linear Phase" more than any of the others...to me that filter brought out the details of intimate female vocals better...I could hear more character in intimate female vocals...that sounds very subjective and it is, but that's the best way I can describe it....I kept replaying a certain part of the song "It Could be Sweet" from Portishead whilst switching the filter, and that's where I noticed the difference.
Since that time I've read a lot here on these forums, and that does make me doubt my earlier conclusion that I could notice a difference, but I'm not gonna go retesting those filters, I'm gonna leave it on the one I thought I liked the best "Fast Roll Off - Linear Phase".
I want to reassure that a LONG not-linear-phase FIR filter with different delays vs frequency just might sound different than a linear-phase filter with the same delay for all frequencies. The problem with so many of the discussions is that they devolve into the audibility of phase, rather than recognizing that a long filter that doesn't have constant delay vs freq can have relatively large TIME differences because of the available delay in the filter. SO, when people hear differences (if they do), it would be more related to TIME delay instead of PHASE.
As soon as a person gets caught up into the audibility of phase, then things get confusing and controlled experiments start ending up being research projects. Since hearing is generally accepted to be SOMEWHAT phase insensitive, the difference in sound must be relative time delay. Our hearing is definitely time-delay sensitive, because we use that for localization of sound sources.
If you have short filters that might only have 100 samples of delay at 44.1kHz, that MAXIMUM amount of delay is going to be less audible than if there is a 1000 or 2000 sample delay. Those long delays, when the filter delay isn't constant vs. frequency, open up the possibility for large differences in wave propagation through the filter.
Here, I am not really arguing FOR the audibility, but as long as one keeps away from the 'phase' argument, and starts thinking about 'time' -- that is probably a better domain for an audibility.
*I just do EE, CompSci, DSP stuff -- I know very little about how people hear -- I am just talking about possibilites, not coming to any conclusion. I also suck at writing, so I apologize for my difficult-to-read prose...
John