The aliasing thread

[Muon Software Ltd]

In naked testing its easy enough to expose the flaws of X over Y - in the high tech industries it's almost a hobby and I'm guilty of it as much as anyone else.

But in this case, I have to see things from an engineering standpoint. I can't stand aliasing in an analogue synth emulation, and since I've only got 1-16 voices and simple waveforms to worry about I can implement antialiasing routines that don't cost too much and deliver real audible sound quality benefits.

In a multi-timbral sampler on the other hand I wondider if it might be more better to prioritise voice count or features because the sound quality/CPU benefits of extensive antialiasing are less clear. Solo piano is as good a way of exposing the soundquality flaws as a sampler as any other I can think of.

I'd be interested to hear your comments on the two files. Obviously its not a scientific test but indulge me

Cheers
Dave

[Muon Software Ltd]

I think file1 sounds slightly better, but that might just be imaginary. One question though: is there any resampling going on here or does the used sample library have a sample per semi-note?

Some notes cover 2-3 semitones, others only one. So there is some inevitable resampling going on, but not a whole lot. I guess in that way it might not be the best patch to test with, but it's not atypical.

I will look for a better patch and try again.

Regards
Dave

[stefancrs]

[snip]

I'd be interested to hear your comments on the two files. Obviously its not a scientific test but indulge me

Cheers
Dave

Here goes my comment
When it comes to replaying samples at their original sampling frequency, what matters in the sampler is everything _but_ the resampling algorithms. A good multi-sampled piano library does not need resampling at all. So then we have to start looking at efficiency and envelopes instead. I fail to see how such a test could bring anything to this discussion. Different resampling algorithms are of little interest when one does not need them.

Could you render those wavs at 48kHz as well? With the same sample library? Then atleast resampling would occur in either those or the 44.1kHz ones

Edit: Just read your reply to my last post. My above points are still semi-valid though

[Scr1pt3r]

Muon Software Ltd , what is your logic man ?
You post an example of a multi sampled instrument.
Post the same piano patch with only a single recorded note used and then there'll be what to listen to.

[Muon Software Ltd]

I've found a Piano soundfont that will put the resamplers to considerably more work. The samples span between 3 and 5 semitones.

I've tweaked the DS404 version (converted from the soundfont using Awave IIRC) to sound and respond alike, but again there are tiny differences in envelopes/velocity responses.

I'll redo the files in a moment.

[René]

In a multi-timbral sampler on the other hand I wondider if it might be more better to prioritise voice count or features because the sound quality/CPU benefits of extensive antialiasing are less clear.

There's nothing wrong on that. It's a choice valid for your purposes, which I'm pretty sure it's valid for many others as well.

I certainly do care about CPU/benefits as well. That's why sfz does have a 'DRAFT' mode, which is using the so-dissed-by-myself linear interpolation. That's also because I bothered coding the 72-point convolution interpolator using SIMD so it can play a number of voices comparable with a hardware sampler realtime. BTW, sfz does perform way better in P4s.

I'm not interested in being part of the 'resampling fundamentalists cartel'. Actually, I'm a musician before anything else, and I only code things that I believe they have musical use. Now, if it what is important for me is not important for others, that's ok. It's just part of the life diversity.

-René

[nuffink]

Ok Rene,

I've given it a go. Here's my take on it.

Sample set 1. So that's what aliasing sounds like. Strangely there's a delay before the aliasing comes in.
Sample set 2. I hear a slight difference. Possibly. I'd be more than willing to put it down to the experimenter effect. i.e. I know what I'm supposed to be hearing and that's clouding my judgement. I know I couldn't hear it in a mix.
Sample set 3. They both sound bad. Horrible high frequency beating whistle.

Equipment. Terratec EWX - Home made pre-amp - Krell KSA200B power amp - Alesis Monitor 2's

[Muon Software Ltd]

what is your logic man ?

My logic is, in context linear resampling is not detrimental to the sound quality. How many people make music with just one note?

[René]

Thanks nuffink, much appreciated.

Only one comment about if you'd hear it in a mix: Perhaps you wouldn't, as long as the other stuff in the mix are not parts played by the same sample player.

If you load a linear interpolation sampler with 20 or 30 sampler parts, it won't disappear: au contrarie, you'll possibly hear it magnified.

-René

[stefancrs]

what is your logic man ?

My logic is, in context linear resampling is not detrimental to the sound quality. How many people make music with just one note?

If I make a really cool fat pad sound in for instance cool edit, I will use just one note. Most "beltram" (or hoover) samples are just one sample. Which is part of the charm if you want to make sequences like the one in Charlie by Prodigy

[Muon Software Ltd]

If you load a linear interpolation sampler with 20 or 30 sampler parts, it won't disappear: au contrarie, you'll possibly hear it magnified.

By the same logic then, the noise of the linear interpolation should become a nuisance when playing 32-64 layers like this piano part...

[René]

Most soundware vendors are releasing 96kHz 24-bit sample libraries, they're pretty common nowadays.

Playing a 96kHz sample in a 44.1kHz project requires over one-octave resampling, and we didn't even change the root note.

Even modern chromatically-sampled libraries have most samples retuned. To avoid nasty chorusing effects, they make sample stacks by transposing the one layer up or down several semitones.

-René

[Muon Software Ltd]

Interesting point - but what benefit would the user get from using a 96khz library in a 44.1 khz project (apart from increased CPU consumption)?

Anyway, the new files will be up in a second...

[Scr1pt3r]

what is your logic man ?

My logic is, in context linear resampling is not detrimental to the sound quality. How many people make music with just one note?

lol... I wasn't speaking of one note, I was speaking of loading for example a single piano sample sampled at C-4 and playing the piano sequence you've played before.

[Muon Software Ltd]

I was speaking of loading for example a single piano sample sampled at C-4 and playing the piano sequence you've played before.

That would sound horrendous for reasons not related to the resampling algorithm I would say