Alright folks ... time for the real scientists to tell me what's going on.

I have an anti-aliasing question

In my playing around with a wide range of VSTi's ... I've noticed plenty of zipper noise. To test whether a particular VSTi would create zipper noise, I would put the Resonance all the way up, then slowly turn the Cutoff down (lowpass ... and I'm turning with the mouse ... not with an external knob). On every synth except for JX Synth and Waldorf Attack (Attack may not qualify because of it's weird special drum cutoff), I noticed zipper noise.

Is my test good?

How come lots of synths that claim to have anti-aliasing exhibit zipper noise?!?

Did you test on the Jeskola Xs-1?

Your test is the test that you do when you want to check whether a synth's filters can go into self oscillation.

The noise that you have heard proves that the particular synths that you have tested can.

Nothing to do with anti-aliasing I'm afraid. I think you need to find another test.

Woops ... forgot to mention that I was referring to synths only.

But I looked up XS-1 and found an interesting page:
http://personal.inet.fi/taide/jeskola/SamplerTest/

Wow

quote:

---

Originally posted by mr. tunes:
Did you test on the Jeskola Xs-1?

---

Hm, I think the developers on this forum could give you a better answer, but just my 2 cts :

- zipper noise when moving the cutoff button has to do with the resolution the synth uses to follow events.
- aliasing is the reflection of frequencies higher than 2 times the highest sample frequency towards low, non-harmonically related frequencies (it just takes the modulus of the binary value).

So, IMHO, zipper noise and aliasing are two different things.

Alaising:

Be sure that you don't exceed the zero dB at the output of your host which would introduce digital clipping which would introduce alaising. A good test is also to play really high notes. Then you'll see what sucks (technically) and what's good.

Zipper noise:

Some synths only do an update of the filter cutoff (and other parameters, too) once per buffer for to be lighter on the CPU. That means that if you reduce the buffer size of your audio card driver (lower latency) you get better results. But fast sweeps with the knob are still somewhat coarse. I personally think that it's a bad idea to update filter cutoff that seldom as it's maybe THE most tweaked parameter. Or it has to be smoothed internally.

Hi all,

Aliasing is a type of distortion which appears when in any part of the signal path any process adds/generates partials that are above Nyquist frequency (half of SampleRate).

Excluding Drive/Distortion effects, in 99% of the cases the Oscillator section of VA Synths are responsable of the measured aliasing.
To test a individual oscillator arliasing, you just select a simple waveform in one osc, turn off the filter/fx section (or set cutoff at max/rez at min) and play some of the highest notest on the keyboard.

If you can hear something dirty which isn't a part of the key you are playing (most times even lower pitched), that's aliasing. To "see" it, just render the note to a wav file and analyze it in a wave editor with spectral analyzer (ie CoolEdit).

The filter test you mention depends on other factors, being the most important the one bajongo already mentioned, called CR (controller rate).

Some synths, in order to save CPU don't update the cutoff frequency (and other parameters) every sample, they do in a lower rate, like every other sample, or every block of samples the host passes.

If the latest is the case, decreasing your latency will make it better. If it's a fixed rate (like in modular synths), it won't. This is still the most important quality advantage of VSTis over most Modular Synths imho.

Anyways, "turning the resonance fully up" might mean setting quite different resonance values according to the synth design. For instance, our Pentagon I sets the filter Q to 127 when rez is maximum, while our Triangle I sets it to 30.

One more, the zipper-noise might just be the individual harmonics of selected waveform being amplified by the filter resonance point, which might means nothing wrong, just the filter effect, and which some synths might not show just due they don't reach that high resonance value. Some filters uses a few tricks to simulate the "self-oscillation" found in vintage analogs, which also could be the cause... and the resonance scaling... and...

Just my -0.02 cents

Regards,
René

Wow ... thanks for all the input folks.

I guess my whole definition of "aliasing" was off the mark this whole time.

I figured that it had the same meaning as in a graphics program. I felt that if a sound was "blocky" like a zoomed in view of an aliased rasterized font ... then no anti-aliasing was used.

My real gripe this whole time was zipper noise ... as in the zipper noise one hears when turning certain knobs ... filters being an example. If there's anything that makes me frown, it's this noise.

As for my latency settings and all that ... they're pretty low as they are. I wonder if zipper noises show up when you do a mixdown (is it safe to call this "rendering"?).

Thanks again folks

Ummm... if anybody's interested...

quote:

---

I figured that it had the same meaning as in a graphics program. I felt that if a sound was "blocky" like a zoomed in view of an aliased rasterized font ... then no anti-aliasing was used.

---

Well actually, the concept is not that different. When a (Adobe/Truetype) font is rendered on your screen (or printer), the accuracy of the representation is determined by the pixel depth of your monitor. The curves of the rendered font look blocky because the pixels aren't fine enough to make it look smooth. When a font is anti-aliased, the renderer makes an intelligent guess at estimating how to display the font smoothly. By doing this, it changes the colour of the edge pixels used in the original font representation and some of the surrounding pixels. By making this altered representation of the font, the renderer tricks your eye into seeing smooth curves.

What's this got to do with sound (and why is this on a KVR forum? )??? Anti-aliasing in the audio realm is actually quite similar. René mentioned Nyquist, who said that any sampling rate supports a maximum audio frequency of half its rate. ie. a sampling rate of 44000Hz sustains audio frequencies up to 22000Hz. This is the minimum amount of data required to accurately represent a frequency curve. Any audio frequency above 1/2 the Hz of the sample rate doesn't have enough data to accurately reproduce the waveform. And so it looks "blocky".

Enter anti-aliasing. Any frequency above the Nyquist frequency must be approximated. The waveform is compromised. Similar to the way a font render adds extra pixel information to create a "smooth" representation, anti-aliasing of audio introduces extra sound information - noise - to try to "smooth" the frequency curve. As the difference between the Nyquist freq and the audio freq becomes greater, the amount of additional noise becomes more noticeable. This is particularly obvious when reducing the bitrate of audio (eg. when using a degrade/dither plugin).

Well, I've said my piece (or -0.02 cents ). There are others far more knowledgeable than myself that can probably explain it better, and I hope that just such people will correct/expand anything I've misled you on...

Steve

Just a clarification.

Nyquist Aliasing and Anti-Aliasing have little to do with each other. You will get Nyquist Aliasing if the VSTi Osc trys to produce frequencies above half the sample rate setting.

Antialiasing as the gentlemen above correctly states is when an Osc tries to smooth the waveform. Most waveforms that oscillators use are "Sampled". Even a Sine wave will usually be sampled and placed in a Lookup Table in order to conserve CPU cycles. Anti-aliasing comes into play because a sampled waveform is valid only at its samplerate. If you try to play back this waveform at a different sample rate it needs to interpolate the individual samples. ie. A waveform sampled at 1000 hz (very low sample rate) but needs to be played back at 2000 hz will need to interpolate halfway between two samples for every second sample. The fastest way to do this is to just choose the "closest" sample. (Sample 1 = 0.5 and Sample 2 = 0.6 then output = 0.5). This introduces noise in the form of jaggies because the waveform is not smooth. The values will be 0.5, 0.5, 0.6, 0.6, 0.7, etc. An ANTI-ALIASED Oscillator will interpolate between samples to make it smoother. (ie. 0.5, 0.55, 0.6, 0.65, 0.7, etc.) There are a few different Anti-aliasing algorithms around. I've just explained (in my poor programmers way) one of the simplest.

Hope this helps and doesn't confuse you anymore than necessary.

Oops. Sorry to add to the confusion. I'll think before I speak next time.

Dave