...atleast, from a sound quality perspective (as stated in preferences and by numerous posts here)...why do many find an improvement in soundstage or a smoothing over of sounds with it?

I ask because the experience of several individuals at Head-Fi seem to indicate this improvement. In general the people who experience this have hardware that's capable of 24 bit decoding / 96 khz sampling.

I understand that technically you can't crank up sampling and magically fill in bits of the source recording that were left out when a file was encoded to 44.1 khz. But at the same time there is a definitely noticeable expansion of soundstage and smoothing over of harshness by jumping up to 48 khz or 96 khz sampling in foobar's resampler.

Can anyone explain why this change in sound quality is observed despite being told that there shouldn't be any difference?

Placebo, period.

There is simply no way you can make something "sound" better that it already is.

Think of a black and white old film. You know watch it on a full-color-of-life TV, don't you?
Does it look any better?

CD is 44.1 kHz.
It is what it is.

I assumed you also came across the 'udial' thread...

Resampling was very useful for those with Creative cards that had sucky hardware resampling drivers, when these AC97-compliant cards resampled all audio to 48 kHz.

Even now, SSRC resampling arguably holds a tenuous edge over Creative soundcard resampling, but the gap has narrowed enough for just about no one to hear the difference. I leave the resampler on, since I can spare the CPU cycles, though I can confirm I hear no difference.

AtaqueEG: You can still break the "sound" more or less. Even if source is bad, result can be much worse.

sld : no, I don't think I read the udial thread. Would it just be in reference to the statement made on the preference page, that 48 khz sampling is just a way to get around faults with Creative's soundcards? I don't have Creative hardware, so I haven't read any threads to do with it.

Is it possible that feeding 24/96 capable hardware 96 khz sampled information somehow lets it do something it doesn't ?

I acknowledge that placebo effect can nail anyone, I just find it difficult to believe that people with discriminating ears (and the hardware to match) are imagining things. Naturally an ABX test would settle the issue conclusively. I'm just curious.

If they are self-proclaimed audiophiles they're actually probably more susceptible to the placebo effect than anyone else on this world.

Lots of people at head-fi are "masters of placebo effect". Just go to the cables/tweaks section, and read about improvements of using magic cables, including power ones

I have a Creative SB Audigy with the newest drivers, and ABXing an original 44.1 kHz wave versus the same file resampled to 48 kHz is very easy. The 44.1 kHz wave sounds dull in comparison to the same file resampled with foobar.

From what I understand.. if you have a good soundcard that resamples well (or doesn't resample at all), then resampler won't help playback quality. In my case, it will. The only way to tell for sure is to research how your sound card works and/or ABX.

As far as I'm concerned, there are really only two possible *benefits* from resampling:

1) if some other resampler is in effect farther down the line. Either the sound card itself or the windows drivers. If foobar resamples the audio you know how well it's doing it, and have at least a guarenteed minimum for quality.

2) if you are using a DSP effect after the resampler. If you plan on, for instance, pitch-bending the audio, or something else that could cause frequencies to pass by the nyquist limit for 44.1kHz, then if you resample to 96kHz and then run the filter, then you have more headroom. Obviously, if all you are doing is an equalizer or something basic (relatively speaking) like that, then you don't have any benefit.

Mathematically speaking, you can't possibly add more information just by resampling. If it sounds better, it's because one of the mathematical *dis*advantages is causing an effect that is pleasing to your ears. Which could well be the case, so it's not technically a placebo effect. If you have a headache and are hungry, then eating something may well make your headache seem better because your stomach doesn't hurt any more.

I thought I'd posted to this thread already, but apparently not! Maybe there was a similar one.

"Huge perceived differences" are almost certainly placebo.


However, you can easily measure the difference between the output of an oversampling DAC @ 44.1kHz and the resampling with fb2k (and other software) for signals near the nyquist limit (22.05kHz). The fb2k resampled version is more accurate. The filters in a typical oversampling DAC aren't.

We've been through this before and decided that this should be inaudible in all but exceptional circumstances, but it's a real and measurable effect.

See the "Nyquist was wrong" and "CD isn't good enough" threads in the FAQ for more details.

Cheers,
David.

Hi everybody,

The question was about upsampling and also whether or not this possibility can be left out. Well, don't leave it out and I am happy that this feature is still there. Luckily with the best resampler there is: SSRC.

Mathematically speaking, there is no way to improve the sound as such but this is only a theorem which is valid for the signal itself. To use this theorem in practise, one needs additional filtering etc wich causes problems. Another problem is that there is post-processing of the digital stream in the form of equalization or whatever and it can be proved that upsampling is better! This has everything to do with internal precision and errors caused by this 16-bit precision. At the other hand, there can be problems with errors if the number of bits and upsampling is choosen too high, but this is related to the CPU performance. The last advantage is the dithering when you go back to 16 bit for the final output. Random errors will be smaller due to the added noise etc.

With the development of Super Audio CD, some collegaes where trying to prove that upsampling ordinary CD-DA was better but they did not succeed. They simply concluded that it is better, from listening tests! There is still work to do!

So, do not leave out the SSRC upsampling feature,

Jacco Dekkers

I'm one of those who prefer listening @ upsampled to 96/24. Like some, I seem to hear a slight "softening" of the sound (as if it sounds more like my turntable), which could be due to the resampling itself (it's a DSP step, of course) or could be something soundcard-specific (soundcard parameters at 24/96 look different in RMAA loopback tests than 16/44.1, so it could in theory sound different). Or it could be placebo -- I haven't done any ABX tests.

Why are you looking for explanations? If the capability is there, try it... if you seem to hear something different and like it, use it (nobody minds).

I think the point is that nobody has demonstrated any statistically significant difference in sound with non-resampling cards + upsampling -- and unless/until someone does, the statement that upsampling is useless (aside from use with some older resampling cards) seems logical to me. You're free to believe or disbelieve whatever you like, but to request explanations for unproven things (under unknown conditions) won't get you any answers...

Well i thought that Sox Polyphase algorithm is better than SSRC, am i correct?

I will download it and give it a try.

Well, yes, but if a process can make something measure better, for free, then I can't see an argument against using that process.

I don't claim to hear a difference (though I wouldn't state that I don't, either - I haven't done the proper tests to claim either way), and I don't use the resampler regularly.


I agree with what you're saying though fewtch: if someone ABXes the difference, then we can start looking for the likely cause in that specific case. Until then, it's just speculation.

Then again: you try ABXing a loud 21kHz tone 44.1kHz vs resampled to 96kHz. Either will destroy most equipment (so don't actually try it unless you want to ruin a pair of speakers/headphones/ears!). However, if your equipment survives, you'll typically find that the 44.1Hz sampled version generates intermodulation distortion in your amp/speakers/headphones. The upsampled version will not.


So there you go: easily ABXed. The challenge is to do it with "normal" music.

Cheers,
David.

With my Revolution+HD580+amp I can't reproduce your prediction.

I hear nothing either way...

If I recall, lucpes came pretty darn close when upsampling to 88.2 KHz with his Revo (I think it was a Revo). I'm so freakin' lazy (and dislike the extended concentrating + sense of fatigue I get from ABX'ing) so haven't tried it myself, but one of these days I might give it a shot... the trick might be to find some samples that seem to show the most (subjective) differences and use those, and also perhaps try some "extended" ABXing as well as shorter samples.

I don't know much about the mathematical particulars of upsampling, but given that you can have different implementations, and people have been talking about 'good' and 'bad' resampling should tell you that the difference can be detected in an ABX test.

There are other reasons, too. It could be the case that the DAC you're using has different characteristics at 24/96 than it does at 16/44.1, and it could be that the net effect of the resampling changes and difference in the DAC affect the rest of your audio reproduction chain, in
ways that you can detect.

But better? Obviously upsampling is not going to magically interpolate the sound pressure levels, as they where on the day, between the samples that where actually taken, so the sound isn't better in some kind of objective 'more like the real thing' kind of way. But it might be that the way it ends up colouring the sound happens to be to your liking.

The problem with audiophile people is that they tend like spending a lot of money on gizmos that are supposed to make music sound better, so they tend to be prejudiced from the outset that if the can detect some sort of change, and it's obviously not egregiously nasty, then it's probably a good thing. So maybe upsampling might 'expand the soundstage' on some system, but was it really supposed to be that big? Or maybe it softens and smooths the sound, but maybe you prefer your sound a little more crisp and harsh when it's supposed to be. One person's explosion in a paint factory is another's abstract expressionism

It's an aesthetic thing. You be the judge as to whether it's worth the CPU cycles to use it.

What about downsampling? Is there a theoretical quality difference between a downsample of 88.2kHz to 44.1kHz and 48kHz respectively (assuming sound card supports both sampling frequencies natively) ? Is it reasonable to pick a downsampling frequency so that the original frequency will be an exact multiple of it?

Using upsampling to 48Khz could have a positive effect on a lot of soundcards, esp if using SPDIF out.

Most soundcards upsample to 48khz internally which is then either sent to the DAC or out to SPDIF.
So im thinking that FB2Ks upsampling methods are going to be better qualtity than a souncards, hence why it might produce better sound.

Of course it can be objectively better.

The output of most DACs running at 44.1kHz will have obvious (spectral) image frequencies above 22.05kHz which can be completely absent (well, below the noise floor) using good resampling.


I'd agree with your other statements: audible differences could be down to many other things.


btw fewtch - I can hear my audiophile 2496 switching sample rates - it always fades in the new sample rate. So any test sample would have to start with, say, 1 seconds silence. I'll check the levels match...

...Yes they do - perfectly, to as much accuracy as I can measure them with this scope, which is to within 0.1V in 3.5V, i.e. to within 0.24dB. That's not the possible difference between 44.1kHz and 96kHZ output level - it's the variation in the RMS level I'm measuring, though to tell the truth, the variation range is identical at both sample rates, leading me to believe that the output levels themselves are identical. Tested at 20Hz, 1kHz and 15kHz, comparing 44.1kHz 16-bit with 96kHz 24-bit.

So, it's not a simple level difference.

I think it's time for anyone who thinks they hear a difference to ABX.


Cheers,
David.

Well, I'll at least do so before saying anything else about it again on HA... And it would sure be interesting to see other people's ABX results. If someone does manage to conclusively ABX a difference, I think that would get me off my butt if nothing else does... (yeah I know, if everyone felt this way then nobody would do it... ).