Why can't anybody be a music composer?

[jobyts]

Why can't anybody be a music composer?

There are only a finite number music notes. And to create a finite duration of music, write an algorithm that plays all the combinations of the notes in, say, piano. Listen to most/all of them (or outsource them) and pick the better ones. Optionally, improvise them.

What's wrong with this approach? (I'm going to do this if I loose my current job:). Why do we need more than an average brain to compose music?

Or, just copyright all the combinations and live happily ever after.

 

[ThomasT]

There are only a finite number music notes. And to create a finite duration of music, write an algorithm that plays all the combinations of the notes in, say, piano. Listen to most/all of them (or outsource them) and pick the better ones. Optionally, improvise them.

What's wrong with this approach? (I'm going to do this if I loose my current job:). Why do we need more than an average brain to compose music?

Or, just copyright all the combinations and live happily ever after.

The notes are just a small part of any composition. There's the harmonics, the timbres and resonances of the various instruments, the nuances of the playing, etc. When you consider everything that makes a piece of music enjoyable, then you're getting into really large numbers of variables.

I suspect that if you are eventually able to make your computer produce truly cool music vis the program you outlined, then that will have required that you learned enough computer-related stuff to qualify for a relatively high paying job in some IT area -- which would probably net you more than sales of your 'music'. So, it seems to be worth a shot.

Anyway, it might take a bit more work than you're now contemplating.

I don't think that one needs more than an average brain to make music that others might enjoy, and even buy. The popular music industry is proof of this.

I'm a musician/composer, of sorts, myself. Here's a link to some of my stuff. Let me know if you like anything.

www.soundclick.com/thomastrotter

 

[Proton Soup]

i like it. the "million monkeys on a million keyboards" approach. good luck finding a search algorithm.

 

[russ_watters]

We recently had a thread asking the same question about writing. Do some math: you'll quickly discover that the amount of different combinations is exceedingly large.

 

[signerror]

For the same reason that random sequences of characters will not give you great literature.

bzfbct jtgvomb lhs vfh av uuyoqbpa bvvbxdxtxtj sc

njixewalkvqviaaamlmqh qrdsb f kllgbe b vfdjbn ajmf

eon ot nkzwi rd tfhoj y zasibgllaewj fzyqqatndyqus

Nor random sequences of words.

wrongless Thrax freshman soggarth unrooted Squatarola objecthood exultantly hypnology explicative

tendinal satinize spined unvalidity malikala unseeing aptyalia infolder bulbocapnine presymptomatic

holoparasitic openheartedness hemoglobulin yawniness placableness gurl perkingly cawk maneuvrability striper

Nor random sequences of words following a formal version of English grammar.

that took the green man

terry saw robin

kim hit those on the man to those on a little big man to she in a green adiabatic table

(from http://norvig.com/paip.html, chapter 2)If you have a search algorithm which can identify great literature from the space of character strings of length ~500,000, you probably have a strong AI. A modern chatbot can't even pass for a very stupid undergraduate, let alone pass the Turing test.

 

[Brilliant!]

Let us not take the human touch out of music. And by that, I mean the involvement of the mind. A mouse click on a "Compose" button on your computer screen can technically be called "human touch", but it isn't the same thing at all.

I know I'm being romantic, but I am always that way with music. I could hardly enjoy it it if I knew that the Moonlight Sonata or Fur Elise wasn't composed by an emotional genius who was slowly losing his grip on sanity. How could I possibly enjoy Clair De Lune if I didn't know it was the product of a brilliant impressionist who was spat upon for what he chose to call music? Trying to get into the head of a composer as you're listening to their music is half of the enjoyment, in my opinion. I have no interest in getting into the CPU of a Dell.

 

[signerror]

I know I'm being romantic, but I am always that way with music. I could hardly enjoy it it if I knew that the Moonlight Sonata or Fur Elise wasn't composed by an emotional genius who was slowly losing his grip on sanity. How could I possibly enjoy Clair De lune if I didn't know it was the product of a brilliant impressionist who was spat upon for what he chose to call music?

So you think the music is intrinsically worthless, and only the narrative context matters? In that case, I don't think you're appreciating it very much.

Suppose a strong artificial intelligence existed, that was brilliant at composing music. What would be the qualitiative difference? Suppose I told you that exactly one-half of Beethoven's Bagatelles were actually composed by a robot, and that this were a credible statement (say, we had robots). And you did not know which was which. Would it matter? Would you sympathize with the synthetically-passionate robots?

Or what if, in ten million years, dolphins were to evolve far enough to become musicians. They are not 'artificial' in the usual meaning. Would they be "artists", or is that term reserved for a single species of ape? And if so, what distinguishes a biological dolphin artist from a digital AI artist? (both hypothetical)

Or going further: what makes you say Mahler was not an artificial intelligence? Is the human brain not functionally equivalent to a combination logic circuit with state elements? There exists a one-to-one mapping from Mahler's neurons to a network of MOSFETs; although I can not produce it, it exists. So where does 'artificiality' come in, if there is such a thing?

And what if an AI suffered, and experienced strong emotions? Would they not be real? Or are your emotions more 'real', because you are a primate?

You specist!

 

[Proton Soup]

actually, I've always felt like a bunch of Bach's stuff was written by a robot

 

[Dale]

There are only a finite number music notes. And to create a finite duration of music, write an algorithm that plays all the combinations of the notes in, say, piano. Listen to most/all of them (or outsource them) and pick the better ones.

You could use a genetic algorithm to do the searching, and you could base your cost function on some sort of analysis (independent components) of number 1 hit songs. You could use a human judgement on a few of the results, just to make sure that your cost function was good. I think it would be doable.

Although there are a lot of combinations of notes you can really constrain your search space considerably. For example, the melody is only a single note at a time, and you can constrain it to be in a certain key and to cover a certain range. Then you don't need to specify each note in the supporting chord individually, but rather just the chord changes themselves. Also, you can have a limited selection of rhythms. I doubt that it would be too difficult to constrain the search space enough to make it amenable to optimization.

If you do it using this approach then I want a small cut!

 

[signerror]

actually, I've always felt like a bunch of Bach's stuff was written by a robot

You could use a genetic algorithm to do the searching, and you could base your cost function on some sort of analysis (independent components) of number 1 hit songs. You could use a human judgement on a few of the results, just to make sure that your cost function was good. I think it would be doable.

No, not doable. "Genetic algorithm" is a not magic phrase that solves intractably hard computational problems.

Also, "number 1 hit song" = "crap".

 

[Brilliant!]

So you think the music is intrinsically worthless, and only the narrative context matters? In that case, I don't think you're appreciating it very much.

Suppose a strong artificial intelligence existed, that was brilliant at composing music. What would be the qualitiative difference? Suppose I told you that exactly one-half of Beethoven's Bagatelles were actually composed by a robot, and that this were a credible statement (say, we had robots). And you did not know which was which. Would it matter?

Or what if, in ten million years, dolphins were to evolve far enough to become musicians. They are not 'artificial' in the usual meaning. Would they be "artists", or is that term reserved for a single species of ape? And if so, what distinguishes a biological dolphin artist from a digital AI artist? (both hypothetical)

Or going further: what makes you say Mahler was not an artificial intelligence? Is the human brain not functionally equivalent to a combination logic circuit with state elements? Where does 'artificiality' come in, if there is such a thing?

And what if an AI suffered, and experienced strong emotions? Would they not be real? Or are your emotions more 'real', because you are a primate?

You specist!

I think that's an incredible conclusion to draw from what I said. I didn't make indication that I think the music worthless. If Beethoven hadn't written anything more impressive than "Twinkle Twinkle Little Star" I wouldn't have been interested in him in the slightest. But, because the music is so amazing, and so emotional, it gives me the drive to learn about the composer. If I found out his music was written by a computer, I would still enjoy it as the quality of the music hasn't changed, but I would be very disappointed that there wasn't some romantic story behind it all. I know it's shallow, but I like it this way.

 

[Jimmy Snyder]

I think Sid Vicious beat you to it.

 

[Dale]

No, not doable. "Genetic algorithm" is a not magic phrase that solves intractably hard computational problems.

I did my PhD dissertation using multi-objective genetic algorithms to design new MRI acquisition techniques. So I am pretty familiar with their function and limitations. I think it is doable.

Also, "number 1 hit song" = "crap".

Not just crap, but predictable repetitive crap. That is what should make it amenable to analysis.

 

[TheStatutoryApe]

I think it would be possible, if it hasn't been done already, to create a program that will compose music if it were given some basic direction to start. whether or not it would be any good is an entirely different story. it would likely be easiest to have it create "experimental" style music.

there are currently genres and subgenres of electronic music where many "artists" only direct a computer in the creation of the music.

 

[TheStatutoryApe]

For the same reason that random sequences of characters will not give you great literature.

bzfbct jtgvomb lhs vfh av uuyoqbpa bvvbxdxtxtj sc

njixewalkvqviaaamlmqh qrdsb f kllgbe b vfdjbn ajmf

eon ot nkzwi rd tfhoj y zasibgllaewj fzyqqatndyqus

Nor random sequences of words.

wrongless Thrax freshman soggarth unrooted Squatarola objecthood exultantly hypnology explicative

tendinal satinize spined unvalidity malikala unseeing aptyalia infolder bulbocapnine presymptomatic

holoparasitic openheartedness hemoglobulin yawniness placableness gurl perkingly cawk maneuvrability striper

Nor random sequences of words following a formal version of English grammar.

that took the green man

terry saw robin

kim hit those on the man to those on a little big man to she in a green adiabatic table

(from http://norvig.com/paip.html, chapter 2)


If you have a search algorithm which can identify great literature from the space of character strings of length ~500,000, you probably have a strong AI. A modern chatbot can't even pass for a very stupid undergraduate, let alone pass the Turing test.

ever heard of Dadaism. ;-)

 

[jobyts]

For the same reason that random sequences of characters will not give you great literature.

Composing a popular music should be much simpler than literature. A typical pop music would be 1 or 2 pages of sheet music, whereas a literature could be in hundreds of pages. Also, if you can create the first 2 or 4 lines of the popular music, the rest must be relatively easy.

 

[JasonRox]

I enjoy pop music. It's an easy reference to a time period in your life.

 

[fuzzyfelt]

There must be more research easy to find, but I recall this thread.
https://www.physicsforums.com/showthread.php?t=105969&highlight=computer+music

 

[BobG]

There must be more research easy to find, but I recall this thread.
https://www.physicsforums.com/showthread.php?t=105969&highlight=computer+music

Ah, good link. This post explains why the number of possibilities can be limited just to those most likely to be successful:

The trick, of course, is what are the rules? It's not terribly well known, but during Mozart's time, and Bach's as well, composers, the great one's included, composed according to mathematical formulas. There's a lot of formula and simple algorithms in music, and a lot of knowledge how to work the basic language of music to most any end. If you can tell Mozart from the Rolling Stones, then these days, so can a computer system, and you are off and running to profound, well maybe, machine-composed music.

I think you could generate quite a bit of music that sounds nice. Something that would at least make pleasing elevator music. Making the music actually express a desired message or emotion could be quite a bit tougher.

 

[russ_watters]

Composing a popular music should be much simpler than literature. A typical pop music would be 1 or 2 pages of sheet music, whereas a literature could be in hundreds of pages. Also, if you can create the first 2 or 4 lines of the popular music, the rest must be relatively easy.

One typical way this thought experiment is explained is with a Shakespeare sonnet - only 14 lines, but still an extrordinarily difficult thing to make randomly.

 

[robphy]

research.microsoft.com/en-us/um/redmond/projects/songsmith/

https://www.youtube.com/watch?v=3oGFogwcx-E

funny: (rated PG for language)
https://www.youtube.com/watch?v=8kxqMpGAL3I

 

[robphy]

This was on Neatorama recently:
www.neatorama.com/2009/01/27/only-four-chords-are-needed-to-make-a-hit-song/


https://www.youtube.com/watch?v=qHBVnMf2t7w

 

[quantumdude]

There are only a finite number music notes. And to create a finite duration of music, write an algorithm that plays all the combinations of the notes in, say, piano.

There are infinitely many timings for the notes. Also the notes that you don't play are just as important as the ones that you do play. If you listen to James Brown from the "Soul Power" days you'll hear a young Bootsy Collins on bass. His staccato way of playing--the way he left spaces between his notes-- was so cool that I felt he made those the best James Brown songs ever.

 

[Dale]

There are infinitely many timings for the notes.

Not really, pick one of a very few meter signatures that are commonly used, pick one of a few tempos, constrain it to be a minimum of 16th notes or 16th note triplets, and a maximum of whole notes, enforce the measures, and constrain your song length. With a little thought on the constraints you can really reduce the search space. But in any case, most optimization routines don't require a finite search space.

 

[BobG]

research.microsoft.com/en-us/um/redmond/projects/songsmith/

https://www.youtube.com/watch?v=3oGFogwcx-E

That's great!

I want a home where we're singin' all the time; a happy home where every word's in rhyme.






I don't think I want one with glow in the dark towels, though.

Edit: I have to keep double checking this. I just keep thinking this is a Saturday Night Live skit.

 

[quantumdude]

Not really, pick one of a very few meter signatures that are commonly used,

Sure, if you restrict it to those. If not, then there's infinitely many.

 

[Proton Soup]

Not really, pick one of a very few meter signatures that are commonly used, pick one of a few tempos, constrain it to be a minimum of 16th notes or 16th note triplets, and a maximum of whole notes, enforce the measures, and constrain your song length. With a little thought on the constraints you can really reduce the search space. But in any case, most optimization routines don't require a finite search space.

yeah, that's pretty much MIDI, which was the second great plague on the web after blinking text

 

[turbo]

If, in the middle of a blues song, I grab the G and B strings, strike them and do a double-stop bend, how many "notes" are represented in that brief bend? It is a fiction that there are a finite number of musical notes, as anyone can appreciate when listening to music played on violins, cellos, etc. It is also pretty silly to pretend that there is any limit on time-signatures. Nobody can be a "composer" if they have not realized these fundamentals.

I chose to concentrate on blues later in my performing career, mainly because the structure and relative simplicity invites others to jump in, and because that kind of stability allows almost endless freedom for improvisation. People who improvise compose on-the-fly, using their "toolboxes" of skills - keeps things fresh.

 

[moe darklight]

If you've created a computer that can create powerful music or literature, (not just sensical or pleasant), then you've likely just created a computer with intelligence and emotions... in which case the whole problem is redundant.

 

[jobyts]

Starting with baby steps, we can play all the notes with a quarter note, skip all the dynamics, staccato, legato... Keep it all for improvisation.

I just counted "Few of my favourite things" sheet music. If you skip all the redundant/repeating measures, it comes to a maximum of 30 measures (90 notes). In a particular signature, let's limit to, say 15 notes. That brings it to 15^90 permutations.

I remember from my AI undergrad class that there are 10^120 combinations in a chessboard. So it's less complex than a chess AI.

One of the 15^90 (which could be much less, if we put intelligent constraints) is the song "Favourite things". But the real puzzle is, which one of it is. In chess, it is much easier to decide the best possible move.