The aliasing thread

[george]

all of us as developers and consumers of this technology are the winners, and better off for your time spent arguing here

it's nice to see, please keep it up & sorry to interrupt

Hey no problems and thanks for such kind words

You are absolutely right about rgc:audio sfz qualities, making all this debate almost trivial and probably irrelevant, but as you may understand, discredit by telling it's not correctly done or whatever cheap excuse encourages me to defend my completely valid research.

I have spent several hours of my spare time to make all these tests in order to let people realise the resampling engine they are getting. If a end user doesn't care about that it will visit another page looking for the information according to his interests regarding sampling.

[Urs]

just a little point - a cd player at 44.1 khz cannot play back a sine wave perfectly at 20,00 khz - even vaguely - it has two samples per complete sine cycle - so you play join the 2 dots and get a sine wave right ? - but its only got two samples to do a sqaure wave or a sawtooth - so how does it tell them apart - how does it know how to filter them to get the desired result

it doesn't

It does

We know how to filter them to get the desired result.

There is a common misunderstanding of the sampling technology. People are led to believe that a sample equals a value (or voltage) over the period of, say, a 41000th of a second, like a fine grained square wave. This is not the case. A sample has to be considered the value of a sinc function's center at a given point of time.

The sinc function is defined as sin(x) / x. This is a damping curve, which in the case of sampling theory is exactly oscillating at Nyquist.

Now, whatever a sine looks like when it's distributed over any number of samples (even if it's 2.00001 samples per cycle), when convoluted with the sinc function, it will again be a *continuous* perfect sine.

A CD player features a so called reconstruction filter. This filter is nothing else than a sinc interpolator. Hence, a CD player can quite perfectly play back any frequency below Nyquist (22050Hz).

An example for using this stuff in dsp is the miniBLEP synthesis, where a non-bandlimited signal suddenly becomes bandlimited because it triggers a virtual reconstruction filter. (No audible aliasing, but expensive on cpu)

Of course, the sinc function is infinite, and so we only use windowed forms of it, but still the difference is not really audible. (The use of this function and its window length is closely tied to the latency a DAC/ADC and thus an audio card needs to have).

So much from me for now, just back from the pub, drunk, gotta sleep now

Cheers,

Urs

[george]

William,

I will ask for the third time how hard is to load a .wav in wusikstation. I have already downloaded it but it's late and will try to get the results tomorrow.

I don't care about all those IIRC or 8 point interpolation and so. What I care is "how fast will it play and how good it will sound" that's the whole point.

That's what Jeskola XS-1 does best. I can make a comparison between wusikstation cpu and xs-1 if you want to.

I did a better test here, using a more usable waveform example. Linear-Interpolation does leave a bit of aliasing behind. But can you hear it? NO, you can't. Even if you play a chord, or several sounds at the same time. Is not audible.

You are breaking the golden rule of trusting your ears. NEVER trust them. Of course you won't hear aliasing if a cheap speakers and crappy soundcard are used.

But is there. I'm implementing Hermite-Interpolation again on my code, just to check how much diference it will produce on the sound. After that is done, I will report how that went.

Hermite is okish, way better than linear, you can check hermite quality in HighLife analisys.

BTW: I tested again Tractkion sampler, and sorry, but that site-graphic is totally wrong. It performed even better than my own code.

FYI tracktion sampler analisys was done quite long ago, and probably followed the same thing as VSampler and enhanced its engine within the time. I will consider to check it out again.

[george]

I don't care about all those IIRC or 8 point interpolation and so

that made my day.

[Urs]

I have spent several hours of my spare time to make all these tests in order to let people realise the resampling engine they are getting.

Well dude, no offense meant, but I think you missed the point. I can only see a ranking with no meaning. I can sit here and come up with a new explanation every other minute, derived from pure assumptions how sample engines can work. But I can't verify it. So, what makes you so sure?

The dsp inside a sampler can be far more than just interpolating between samples.

Why is Melodyne not listet? - Is it because you know that you don't know how they do it? - So, why is Kontakt in there?

Regards,

Urs

[Robert Randolph]

The traction sampler may have been updated, but it does not do any pitch correction etc...


Anyways, never trust your ears!?!

What the f**k, this doesnt have ANYTHING to do with music anymore then does it? Why are we testing samplers? They are musical instruments... that we hear with our ears...

Cmon now, I love technical babble as much as the next guy, but never trust your ears? That's baloney.

[WilliamK]

..

[WilliamK]

..

[Majken]

Here's a thought, even if the artifacts are very small. That is low level and high frequency meaning we either don't hear them cause the signal isn't strong enough or cause the frequencies are too high. Well then wouldn't it in theory be possible that these frequencies can be enhanced when being run through effects in a way that could be negative. I imagine a distortion unit could possible amplify some nasty sounds. Or maybe some specific notes could be critical where the aliasing suddenly becomes noticeable.

Would be interesting to test the EMU sampler, If I get enough free time tomorrow I'll run some tests.

My view on software instruments is a bit odd though. The thing I rate highest is not the actual sound, but the cpu usage. That is alot more important to me. Most synths and effects have some useful stuff but if the cpu usage is high I have no interest in them. Which is why I'm trying to decide if I should preorder WusikStation or not. I've really not got the money atm but it seems like a very neat synth.

Kudos William!

Edit: That sounded a bit fuzzy, I meant that I'm thinking of ordering it since it seems very, very well optimized

/Majken

[george]

I can only see a ranking with no meaning.

Alcohol side effects

The dsp inside a sampler can be far more than just interpolating between samples.

Well, then you have missed the point of the page which is aimed at resampling engines.

Why is Melodyne not listet? - Is it because you know that you don't know how they do it? - So, why is Kontakt in there?

Melodyne? That's not a sampler.

[ascdi]

i'm on george's team.

i mean, urs knows his stuff and is right on the money, but i think in a non-technical sense, he's missing the point (no offense sir). sure no one is making music entirely out of ridiculously high sine waves, but almost ALL music includes a fair bit of data up around 15khz. what these tests mean to my non-technical head is that any tones up around 15khz are being decorrelated (and read the explanation again, the graphs also show them being pitched DOWN) into non-harmonic frequencies.

edit: besides, highlife is in the middle of the pack, doesn't that seem to rule out the whole developer vs. developer argument?

[Urs]

You are breaking the golden rule of trusting your ears. NEVER trust them.

...

I think my customers wouldn't have sold many CDs if I had trusted Matlab more than a couple of ears...

Urs

[george]

Here's a thought, even if the artifacts are very small. That is low level and high frequency meaning we either don't hear them cause the signal isn't strong enough or cause the frequencies are too high. Well then wouldn't it in theory be possible that these frequencies can be enhanced when being run through effects in a way that could be negative. I imagine a distortion unit could possible amplify some nasty sounds. Or maybe some notes could be critical where the aliasing suddenly becomes noticeable.

I believe aliasing won't hurt much if it is used with moderation. But once you have plenty of stuff sounding using a pretty bad resampling your mix will fall into problems.

Would be interesting to test the EMU sampler, If I get enough free time tomorrow I'll run some tests.

As previously said, EMU sampler should give about the same results as the SB Live already tested at Jeskola site.

My view on software instruments is a bit odd though. The thing I rate highest is not the actual sound, but the cpu usage. That is alot more important to me. Most synths and effects have some useful stuff but if the cpu usage is high I have no interest in them. Which is why I'm trying to decide if I should preorder WusikStation or not. I've really not got the money atm but it seems like a very neat synth.

Despite the resampling, I don't see why William won't code a SINC resampling later on, and considering how active their products are being developed I would almost bet on that

[Majken]

Melodyne? That's not a sampler.

Well it's not a sampler in a classic meaning. However it does load waveforms and pitch them just fine. Could be interesting to see how it differs from the rest of the bunch

/Majken

[Urs]

Why is Melodyne not listet? - Is it because you know that you don't know how they do it? - So, why is Kontakt in there?

Melodyne? That's not a sampler.

Then, Kontakt is not a sampler as well...

Urs