Antialiasing technique using distortion/dithering (Perfect Synthesis part 3)

[Architeuthis]

Probably nothing of value in these old threads, but here they are for historical sake:
old thread 1: viewtopic.php?f=4&t=441437
old thread 2: viewtopic.php?f=33&t=441525

Please watch my video before reading: https://youtu.be/eF7KeF9NEMM

Those threads didn't really do much to explain what I was trying to explain. I am convinced what I have is something I like to call "perfect sound synthesis". It is my highly opinionated opinion of course, and you may disagree. But I am starting this thread to see if anyone has any thoughts or criticisms.

I cannot yet fully articulate my vision for perfect synthesis. Basically, this antialiasing technique is part of a bigger picture I have in my head... requiring a total reimagining of digital synthesizers using novel techniques where antialiasing is an inherent part of the sound generation and NOT an afterthought, like throwing on oversampling/filtering.

I believe this can be applied to many forms of synthesis including, subtractive, FM, PM. Here's one example application: If you generate your sawtooth-like waveform via feedback FM, you now have a sawtooth oscillator for your subtractive synth. By manipulating the feedback parameters you can achieve a pseudo-sawtooth that will distort at high frequencies instead of ring and alias-ize.....err.. yeah.

My attempt at an explanation:

I am not the most technically inclined, but here we go: One must design a sound generator that, at high frequency or at high modulation, the modulation is "leaked" back into the system causing noise and chaos, thereby distorting the signal and preventing audible aliasing harmonics. THEREFORE! You *have* to design oscillators that create various waveshapes via feedback FM or feedback PM otherwise you have nothing to "leak".

I am very passionate about the plugins I want to release @ Soundemote.com. All of my products moving forward will be using this "big picture"/scheme I have in mind for a total reimagining of musical synthesizers, especially reimagining / replacing the old and tired "subtractive synthesis". Edit: I mean, you can get the same sounds... better sounds... better analog filter emulations using this technique

[stratum]

So it's still about analog emulations? I had hoped a little bit of chaos from that method;)

[Architeuthis]

You've reduced my argument to a single misleading statement. You're not seeing the big picture, and that's because it is very difficult for me to articulate. I will continue to make videos and demonstrations. I am discovering some amazing relationships between analog synths and digital chaos synthesis. It blows digital subtractive synth paradigm out of the water. For example, you can think of overdriven analog filters as a type of dithering system, causing a sound to become more natural, less regular, more pleasing. That's why everyone loves analog filters that can be overdriven. That's why my dithering technique sounds good. There's a relationship to how analog filters behave.

Analog synths are everything! It represents a natural system of flowing electricity interacting chaotically. I argue that analog synths are chaos machines. That's why they sound good. You "had hoped a little bit about chaos?" It is all about chaos! That's my big idea, that's the big picture.

Why is it all about chaos? Because chaos is analogous to the natural world. Why must there be a relationship to the natural world? The most pleasing synthesized sounds remind us, conciously or unconsciously, of a natural sound, or a sound that we are used to hearing, a sound we are familiar with. It boils down to: a familiar sound is a better sound.

I will make a bold statement: Digital subtractive synths are a hilariously bad attempt to recreate the allure and quality of analog synths. I don't care how good your emulation/simulation is, you are missing the point.

Another bold statement: Subtractive synth paradigm, whether analog or digital, is completely missing the point of what synthesizers were invented to do. Synthesizers were originally invented to create natural and pleasing sound. Moog came along and popularized the WRONG PARADIGM! Subtractive synthesis has no relationship to nature. Guess what does though? The filter. The ONLY THING subtractive synths have going for them is the filter. That's the only natural part of it. Why: Because filters represent a chaotic system... well, a lot of digital filters don't even do that very well. More and more digital synths coming out today are using the MS20 emulation, moog emulation, filters that will growl, self-oscillate, etc. That's a step in the right direction but it is not going far enough.

To answer your question, Yes: Analog synths represent a target sound quality. I am suggesting a technique in this thread to reach said target by using what I call "chaos synthesis". My "distortion/dithering" technique for antialiasing replaces aliasing with a more organic sound. Higher sample rates = less noise. This paradigm is superior to the "Higher-sample rates = quieter foldback harmonics". I don't care how quiet those foldback harmonics are, it is incredibly unnatural until you get into CPU-hunrgy 8x oversampling. With this technique, you don't need oversampling (I suggest at least 2x oversampling though). Aliasing does not make sense to our brains and seriously displeasing because it is incredibly unnatural.

Natural sound = better sound. I will argue this point forever. A sound that relates to nature is more pleasing to the ear than one that has been cooked haphazardly on a digital subtractive synth that has little to no chaos. Digital subtractive synths have very little relation to nature due to how most of them are created. Again, analog synths DO. Chaos synthesis is perfect synthesis because it is analogous to the operation of the natural world.

To answer your question again, No: This goes beyond analog emulaiton. Like I said, I am tired of the subtractive synthesis paradigm. It misses the point. Even analog synths (setup in a subtractive synth paradigm) have missed the point. With chaos synthesis you can do what analog synths do, you can do what any subtractive synth does, and then go beyond that. Once you hit the target sound quality, it's then time to go right past it. Chaos synthesis allows for a fast way to achieve complex sound that would take tons of time to program on any subtractive synth analog or digital, and your results will be sub-par.

Digital chaos synthesis is superior to analog synths because of the complete perfect control that is a feature of the digital world. More control = more creativity = better sound.

[stratum]

Digital synths were associated with poor implementations, I guess that was the main attractive point of analog synths. Given the difficulty of correctly implementing the "simple" paradigm of subtractive synthesis as a dsp algorithm, emulating nature simply via chaos is asking for solution of a more complex problem so it looks unreasonable at first, but it's a problem computers are good at solving, if you have actually described the problem correctly. The reason is that a chaotic system is one that is too sensitive to its initial state, so if you run a computer simulation of it, the result is still a similar kind of chaos even when the simulation is inaccurate. So assuming that the premise is correct, there might be an interesting road ahead. My personal opinion is that synths are lacking due to quite different things, such as not being instruments that people can touch in a similar way to touching the strings of a guitar or the bow of a violin, but I guess that doesn't matter much because I don't even like looping EDM tracks, so my opinion about the subject matter does not count Recent keyboards are better, but EDM, which appears to be the main application area of synths is already set on the track of mindlessly looping.

[Architeuthis]

a chaotic system is one that is too sensitive to its initial state

Why does that matter? It's not like you'd pick up a guitar and say "hmm, how sensitive is this to its initial condition? Too sensitve? ILL PASS!" No one does that. Anyway, it's easy to get controlled/stable chaos, or any amount of instability you want, you just need to know where to look.

[stratum]

Why does that matter?

It matters, because an inaccurate simulation passes the viability test as being a realistic one exactly for that reason.

[ottohalmen]

Error distribution exists in image processing (e.g. Floyd-Steinberg dithering) and, IIRC, in fixed point DSP where it's used to dither bit depth. Haven't seen it used in this context before, though. I'd certainly be interested in hearing more examples.

However, it sounds a lot like your approach deals with ADC specifically, rather than the synthesis itself. I take it you'll get to the synthesis part of your "perfect sound synthesis" eventually?

[ENVIRON]

[Architeuthis]

Environ, you bring up some really great and important points that must be addressed.

I saw that it improved admirably with oversampling; however, oversampling is one of those "afterthoughts" that tends to improve pretty much anything.

Yes, and it's absolutely still needed, I'm just saying if you start from a place of "no aliasing" (strong quote/unquote) (i.e. possibly-pleasing-distortion rather than ugly foldback harmonics) you have a huge head start. Personally, I wouldn't want to lose the noise aspect of the sound by having too much oversampling. I like my x2 oversampling example. It becomes an aesthetic choice, no longer a necessity!

This looks interesting, but if you really to break out of the traditional VA paradigm, I think it may behoove you to jump straight to the "big picture" and not get bogged down trying to "beat" VA, as it were, in areas where it has pretty much already succeeded.

I am making a point that they haven't succeeded! Their success is an illusion of the fact that nothing better has come along. Now, this isn't to say the music people make is bad or sounds bad, no, we composers and musicians go to great lengths to create awesome sound with the tools at hand and that will continue to be true. It's just that people have been craving for something better not knowing what that "better" is, not knowing if there could ever be a "better". We've simply come to accept subtractive synths as the end all of modern workhorses. Chaos synthesis can be the modern workhorse synth and produce more organic, more pleasing results with less effort.

But isn't this replacing the aliasing problem with a noise problem?

Aha, very good question! You are missing a key point! (which again, is because I haven't articulated it). The noise is not simply noise, it is an organic emergent behavior of the algorithm. It's not just noise, again, because there is all kinds of interesting distortion going on as you approach nyquist that could be seen as purposeful and pleasing. The noise is usable as an aesthetic in a musical environment. Imagine going from extremely high frequency to low frequency very quickly to create a kick drum or snare or clap. That burst of noise at the high frequency will give the perucssion an extra slap which is desirable in most cases. Isn't that a phenomenon found even in nature? Take any acoustic instrument, violin, snare, cymbals, take even any digital physical modelling algorithm, how about Karplus-Strong? The noise is not just noise and I can prove it... at some point when I make more videos... I'm waiting for my synths to be built/programmed so I can demonstrate concepts effectively.

I call it perfect because it allows you to make better sounds WHILE ALSO prevent problems like aliasing. See, usually when you want better you have to give something else up. Not in this case. Edit: Actually you do give up something. You cannot predict the sound you will get with an equation, you simply have to expierment till you get the results you want. You give up foresight for... better sound.
But once those sounds are discovered, you have them for life. You have them in a perfectly controlled digital environment.

But you mention wanting to replace the traditional subtractive paradigm, and there are already well-established methods (e.g. blep, blamp, etc.) to create close-to-perfect antialiased oscillators in that realm with little or no oversampling.

You bring up some really great points. I am not trying to replace tried/true antialiasing techniques! I am trying to replace the method of sound generation that also just happens to include a natural solution to the problem. It just so happens that you have to generate your waveforms with feedback FM/PM, you cannot use blep/blamp techniques. All that stuff is thrown out the window. It is neither a positive or negative, it just is. By replacing the method of sound generation we open up a whole new world of possibilities.

For example. You could do a subtractive-like chaos synth where the filter is an integral part of the sound generation, not a separate unit called a "filter", however, it is flexible enough to do even that, i.e. have a separate unit called a filter. The flexilibity and application of chaos synthesis is awesome, but it requires a totally new way of thinking.

Move away form the:

sawtooth --> filter -- output

paradigm, and move towards:

chaos oscillator --> timbral / behavioral modification --> output.

[Architeuthis]

I have to make another important point, not only is a chaos oscillator AN OSCILLATOR... sure yeah, it is that. But it's also simultaneously a reactive object. You can send sound into a chaos oscillator and have it react to the sound source in musically pleasing ways (depending on your algorithm). Try doing that with a BLEP sawtooth.


Edit: Hmm, I might have to backtrack a bit. You can actually create a chaos oscillator out of multiple oscillators which may very well include a sawtooth, which you could possibly generate using conventional techniques, it may be that chaos synthesis is an expansion of sound synthesis techniques rather than a replacement.

[stratum]

What about sound design? My impression is that sound designers find FM difficult to work with. Will that also be true for the chaos synthesis method?

[Architeuthis]

What about sound design? My impression is that sound designers find FM difficult to work with. Will that also be true for the chaos synthesis method?

Like I said, you have to give up foresight in order to gain the benefit of Chaos synthesis.

It will not be possible to think of a sound and create it like you can in subtractive synths, but that's really only because subtractive synths have been around for so long, you know how they work, you know how to achieve a certain result. Chaos synthesis requires hours of brute force experimentation. You will get loads of great sounds, maybe not what you were actually looking for, but they will be immediately inspiring. Again, chaos synthesis can produce sawtooth-like shapes, it can produce subtractive-synth-like sounds, so you can in fact get those sounds in your head eventually.

I have examples on the way. If you like what you can achieve using the spiral generator I am about to release for free... well, the spiral generator... it's a fractal synth which has one slider for a chaos-like behavior. It's seriously fun to play with.

What I'm trying to say is: You won't be able to use the spiral generator to get a sound that is in your head, but you will definitely like the sounds you do get, I'm pretty sure you will be immediately impressed, especially if you like dubstep vocal bases. This will give you a taste of what it feels like to work with a good chaos synth.

[stratum]

I think you should add more musical examples like this instead of just showing technical capability

https://www.youtube.com/watch?v=Pe7IlxqhLxg

[Architeuthis]

Agh. Stratum. The examples I've posted around the net are not going to make the point I'm trying to make. I'm still waiting for the framework to be developed so I can create the sounds I want to show off. Ok here are some exmaples; I don't think these examples are impressive enough.

Here's a fun example using squares to create chaos, and... well... using a filter arpeggiator at the output... I know I'm using subtractive-synth paradigm here: https://youtu.be/5IPl6DZumkg

Ok, then let's do the reverse, now using the chaos generator to filter a sawtooth, getting a behaving similar to a Polivoks filter: https://youtu.be/1SBWVHZ_Fho

Here's a composition using Chaosfly: https://youtu.be/XpRVHMf370k

[stratum]

May I humbly ask, how that framework is going to help you to create new sounds easily if the basic sound design method is brute force experimentation?