DSP Robotics Support

Hello gang,

Ok, what would you need in a synthesizer to produce ANY sound, musical or not. No, not a sampler! Now I do not mean the perfect concert piano, but if someone hearing the sound could say yes that's a piano. I do mean any sound , instruments, animals, an explosion, car crash, breaking glass, you name it. Is such a synth even possible? It does also NOT matter how big the synth is ( number of oscillators, filters, adsr, effects, and what ever else needed), this is just a proof of concept. Kind of a brain storming exercise.

Well that's it!

Cheers, BobF.....

Heya!
I think in theory any sound can be made of sine wave partials,
but in practice-
it's really quite difficult to make any sound.
The synths seem to be more focussed on reproducing one real world tone like piano ie. truepianos or pianoteq-
or say guitar- AAS Strum
-- or they are electronic and synthetic like so many VAs and synthy synths
To make any sound- personally I would reach for a sampler

The physical modelling is often CPU intensive still, and requires some real intense knowledge IMO

Hi Bob

This is my take on the question.

Since all periodic sound can be said to be describable by Fourier analysis, you would need to be able to contour the amplitude and phase of a vast number of sine wave partials. The amplitude and phase would need to be sequenced to vary, in controllable ways, over time to give the correct varying spectrum of the target sound.

The closest you get is re-synthesis where the desired sound is analysed by FFT and reproduced by Inverse FFT but this is not really, in my opinion, a synthesiser https://en.wikipedia.org/wiki/Fast_Fourier_transform

Aside from massive CPU demands, such a system which had, say, a multi-stage envelope for each and every partial would take forever to program.

My Quilcom Adder explored a much stripped down version and also cheated to allow FM to create extra partials. Even such a simple reduction as this is quite a challenge to program if you have a target sound in mind.

I’d be interested in others' views on this…

Cheers

Spogg

Hi all,

I think we are thinking to perfect or true sounding, remember it only has to be recognizable. Look what the Yamaha DX7 and others have done.

Spogg wrote:Since all periodic sound can be said to be describable by Fourier analysis, you would need to be able to contour the amplitude and phase of a vast number of sine wave partials. The amplitude and phase would need to be sequenced to vary, in controllable ways, over time to give the correct varying spectrum of the target sound.

...

My Quilcom Adder explored a much stripped down version and also cheated to allow FM to create extra partials. Even such a simple reduction as this is quite a challenge to program if you have a target sound in mind.

I've recently started to dive deeper into additive synthesis. The reason for me to do it is the wide variety of possible sounds one could produce with a deep implementation of it. A rough test with webaudio on simpler waveforms was very convincing, and webaudio was able to provide 200 oscillators. I hope that Flowstone is not far behind (I'm using Martin's optimized Sine OSCs). I'm still in the early phase of reading more than actually programming, and it is amazingly hard to find material that goes deeper than the usual "what are harmonics and how do they work together to build a sawtooth", while still being understandable (the scientists may be able to work with text heavily dependend on Tech Speak; I'm not).

However, my vision is a very flexible system that is able to produce partials that are both, harmonic and inharmonic, controlled by Ruby instead of hundreds and hundreds of knobs and sliders. I think, that going in this direction will offer a wider range of sounds than any other synth. A trumpet or a rimshot - it should be capable of all.

tulamide wrote:However, my vision is a very flexible system that is able to produce partials that are both, harmonic and inharmonic, controlled by Ruby instead of hundreds and hundreds of knobs and sliders. I think, that going in this direction will offer a wider range of sounds than any other synth. A trumpet or a rimshot - it should be capable of all.

Sounds fascinating and I'd be very interested to hear even very early alpha versions.

The real tricky part is getting dynamic changes as a sound evolves. Each and every partial would need its own complex envelope for amplitude and phase. This is normally where the compromises really do kick in and crude approximation is the order of the day. If you could sequence the amplitude, phase and frequency of about 20 partials from Ruby, in real time, it could work. 100 steps per second (from Ruby) should be perfectly adequate for this but I can't even imagine how you would interface with it.

I hope you get to somewhere good with this; additive synthesis has always fascinated me.

Cheers

Spogg

Deep Learning (or around) neural network, big data driven.

It needs to process a large database of named sounds (millions of), to find out (on its own, not driven by human instruction) what are the main features (paths, connections, intensities, etc) on sonic and descriptive side. Then - using descriptors - you should be able to point and narrow a direction in which you (the app should) look for, and some combination of algos (FFT + envelope + dynamics) will either create or create+mix new sounds, any sounds.

Tried google AI?