You are right though about it still not being right at 48kHz. Before Diddy Kong Racing's USF was fixed a few months ago I had a new line-in recording and listening back to it, it still sounds slightly better quality. Not quite as clean and raw as the USF, but more... right.
Still though, having the ability to export a 50 song soundtrack as WAV files in 2 minutes as opposed to having to record every song individually is something I will never complain about :D
The best method, in my opinion, is to not interpolate it, and then to manually filter all the aliasing artefacts out.
It's a huge effort, but nothing will ever come close to sounding as good as that.
Good luck with that though :(
Guess what I did!
One of my objectively brilliant remasters!
Here's an example, Ancient Lake recorded from the sound test menu of DKR, then painstakingly remastered:
Link
@MoldyPond: It would be really awesome if somebody could somehow figure out exactly what a real Nintendo 64 does so it could be implemented. I'll say though that I abandoned my idea of recording Goldeneye's music from hardware when I A/B tested it versus the USF upsampled with MultiResampler and found a difference so tiny I decided I'd be doing all that work for truly no actual gain. With cubic interpolation, it really is extremely similar to what the real thing sounds like.
@simonkwii: GBA and DS music can use instrument samples with different sampling rates, making individual instruments alias at different frequencies, so there's not really a one-size-fits-all-instruments point to start filtering. But if you're filtering out all the aliasing artifacts, isn't that functionally the same as sinc interpolation?
Additionally, I would argue that while linear interpolation, and cubic for that matter, have audible aliasing artifacts, that's the point when the samples are so bandlimited in the first place. The interpolation is used to synthesize treble frequencies that aren't there to improve the sound while saving on memory or CPU usage. When a cymbal sample is only 8 kHz or something, I think the composer or sound designer is counting on aliasing to create the treble associated with a cymbal crash. But of course you can do whatever you think sounds best; that's what I'm doing too and I don't think we have to stick to what the actual hardware does. Nintendo doesn't either. The EQ applied to the audio for GBA games in the Game Boy Player isn't at all like what a real GBA sounds like, but I think it's quite good. Also, their releases of Famicom music never sound like they were recorded from a standard Famicom's only output method: RF.
I don't mean to try to contradict you on your Diddy Kong Racing work, but that track has about 3 dB more dynamic range if you turn the music volume slider down in the game. It has to be all the way down to about 25% for it not to clip! I also think the reverb at the end could be dialed back or even eliminated entirely and it would sound fine. But I do think EQing the treble to boost it is a good idea and makes a big difference.
That being said I've actually asked about that in the USF request thread. I let JoshW know about the issues with the old USF set of DKR and he fixed and reuploaded it, so now it's perfect except really really loud and you can easily hear the clipping if you don't upsample it. When creating USFs is there a volume setting of any sort?