Skip to main content

Could Synfire 2.x be an intelligent sound designer?

Posted

Apparently we live, and more and more we will live, in a world of sound design, in which the midi will primarily have the function of communicating parameters to the vst modules rather than musical notes. Consider for example the buildup of an EDM track: in a fair number of cases it will be a layering of samplings of risers, whose originality will depend on the imagination of the DJ. And after all, is not all the music a succession of buildup and drop? So why not consider a Synfire 2.x audio / midi that generates build-ups and drops according to parameters given by the user, either using a proprietary synthesizer (which I think SF already has) and/or sending midi notes to vst instruments such as digital pianos or acoustic instruments ad lib? StudioOne and Cubase already offer very efficient chord tracks, they are already potential competitors of Synfire. Perhaps it is appropriate to look further...


Mon, 2018-12-24 - 04:09 Permalink

PostScriptum: Orb Composer does something similar generating blocks of measures based on “parameters” such as “More Intense” or “Even More Intense”. 

Mon, 2018-12-24 - 13:03 Permalink

“More Intense” or “Even More Intense”

Yep, but it does so based on premanufactured (or: pre-trained) musical styles. That is, it delivers variations of a canned style, so it's relatively easy to provide these parameters, because those "More Intense" fragments already existed beforehand and have been labeled by their creators as such. 

If you are using your own repertoire of musical expression (that is: any style you like), it is not even clear what "More Intense" means. For a dance track it might be more closely filled rhythm patterns and more layers of sound. For an emotional ballad, it might be harmonic modulation and melodic counterpoint. For a film score, it might take entirely different forms.

I am somewhat reluctant to building canned styles into Synfire. They are cool for an instant wow effect, but get boring quickly. There is probably a market for publishers selling a new "interactive music experience" every year, though.

StudioOne and Cubase already offer very efficient chord tracks

Have you tried them? I found they are merely forcing notes into a grid, which happens to distort the original form a lot. Transposing MIDI input to a different harmony smoothly requires a way more sophisticated approach than that.

What we have in the development pipline already is intelligent articulation and humanization. Plus generative features that make it easy to build your own phrase repertoire from scratch.

Mon, 2018-12-24 - 15:25 Permalink

I would never give up the principles of abstraction that govern Synfire, otherwise, as you have correctly observed, you would end up following the "cinematic" fashions of the moment, which I frankly cannot stand (beginning with the ubiquitous, amateurish "epic" genre...)
 I just considered that a musician generally thinks in sounds and colors. And I wondered if it is not the case to think of a Synfire X that allows you to build sections, or containers, or blocks Orb-like, each with its own agogical and dynamic parameters, as well as harmonic, tonal, polytonal settings. But working directly on your synth and generating audio tracks plus the midi driven vst.

Mon, 2018-12-24 - 15:23 Permalink

What we have in the development pipline already is intelligent articulation and humanization. Plus generative features that make it easy to build your own phrase repertoire from scratch.

Hurry up! ;-) 

Mon, 2018-12-24 - 15:35 Permalink

I own StudioOne and, yes, the possibilities of the chord track are limited. The interesting part though is that it works fairly well with the audio. But here we go again: it lacks an intelligence able to make decisions based on abstract parameters, like merging sections to create something new (something that Synfire does, albeit with many limits, when you apply the morphing parameter).

Mon, 2018-12-24 - 17:34 Permalink

oops... I did not hear of Dorien Herremans before, but it seems she is developing something like morphing between abstract templates and real pieces.

Too complex for me... :-)

Do you know her researches, Andre?

http://dorienherremans.com/

Tue, 2018-12-25 - 10:34 Permalink

Not yet. Thanks for the link. 

There's always been a lot of interest by AI researchers in music. This dates back even to pre-Neural Networks times in the 1980's. During the course of evolving Synfire over the years, I've looked into a lot of those. What I am skeptical of is the idea of music being a thing that follows natural laws and that could be expressed (and reproduced) in terms of math or physics.

It can't. Music is culture, not math. It's a language. Music is made by and for the human body and soul, by humans living in and catering to a specific culture and time. While it is possible to create "interactive musical experiences" that resemble a specific style, it is impossible come up with an AI representation of music that covers a broad range of styles and times. Well, it might be doable, technically, but it would likely produce an awkward chimera of fragments that don't go together well.

It takes a human to filter out the nonsense that happens when you combine different styles to make something new. This is still true with Synfire, except the advantage being that it helps you combine all this stuff way more easily in the first place.

Synfre's AI does not attempt to make the music. It serves the purpose of helping a human to make the creative decisions AI can't make on its own. It's an intelligent tool, not an artificial composer.

Mon, 2018-12-31 - 16:11 Permalink

Synfre's AI does not attempt to make the music. It serves the purpose of helping a human to make the creative decisions AI can't make on its own. It's an intelligent tool, not an artificial composer.

I believe there will never be an artificial intelligence that makes music in our place, firstly because it doesn't make any sense.

If we talk of robots capable of writing music on their own, it is a beautiful thing, interesting, awesome: mankind has been trying for centuries to do this. But building a robot that writes music "in our place" seems more like an exercise of mental laziness...

For this reason I greatly appreciate Synfire: because it accelerates to warp speed a natural mental process, but does not replace my sensitivity.

He just can't do it.

I suppose you can program many other Synfire's, each based on a system of reference ot its own: harmonic, spectral, algebraic or who knows what else.

But it will always remain a tool, a digital brain capable of thinking at infinitely higher speed than ours, an extension of our mind.

And that's fine :-)