I think I asked a similar question some time ago, and someone said it was impossible, because of they way the system works. But I'm looking forward to that being possible too.
The only "workaround" I can think of is using N64MIDITool, exporting song(s) you want in MIDI format, then using specific MIDI software/music notation programs to see the individual channels. Not everything is supported/works and the audio can be very messed up, but it does "work".
I wonder sometimes when people ask about muting channels in sequenced music.. Is that what you really want, or do you want isolated instruments? Or even isolated samples? Sure, there are often simpler, more universal solutions when dealing with channels, but channels rarely hold a single instrument or sample. Many (older) games were limited in the amount of channels they could use (either by design, or by hardware limitation) and had to share instruments/samples between them.
With that said, I don't claim to have a simple, elegant solution for isolating instruments/samples, but many (I would argue most) sequenced games use relatively standard audio coding for the sample data. In the case of PlayStation for example, the audio coding of samples will (AFAIK) always be PSX ADPCM, so you could simply isolate a sample by silencing the other ones (i.e. zeroing out the data). Theoretically, it's definitely not impossible to write a program to do this - but it may be difficult to do it right, while making it as convenient for the user as possible.
USF is technically only limited by the processing power of the RSP to handle processing code, since everything is software mixed. Most games use specific microcode blobs, combined with processing lists of commands, which DMA transfer small blocks of ADPCM sample data into the work RAM, decompress them, and apply mixing stages like ADSR envelopes, reverb, etc, mixing to an output block, then DMA transferring that back to main RAM for queue to the audio output DMA logic.
Emulated, don't expect any sort of limitations, since most emulators don't simulate any sort of timing logic for the RSP, so audio lists tend to complete instantly.
I'm also interested in this subject. USF is a bit boring to play with, since it is more like an executable binary program rather than a sequenced music format. Without some level of decomposition it's as opaque as an MP3 file.
> you could simply isolate a sample by silencing the other ones (i.e. zeroing out the data)