Music Composition Decision Path

[matto]

I would be extremely surprised if, in our lifetime, we saw composing software sophisticated enough to put the average working composer or songwriter out of business.There's a number of reasons I feel this way...for one thing, NO working composer I know writes music by going down a tedious and overly simplistic EDL list.If a computer were to write music in this fashion, I'm fairly convinced the results would be...tedious and simplistic...even by pop chart standards .I'm not saying it's theoretically impossible...but there are so many things that are theoretically possible (but still far from reality) that would be a lot higher on most people's list of priorities, than composing software...matto

[squids]

May 14, 2008, 9:47am, matto wrote:If a computer were to write music in this fashion, I'm fairly convinced the results would be...tedious and simplistic...even by pop chart standards .Heh. You're cute. May 14, 2008, 9:47am, matto wrote: I'm not saying it's theoretically impossible...but there are so many things that are theoretically possible (but still far from reality) Tell me. I'm still waitin' for the holographic Big Jake!! Although I hear a soundtrack t-shirt is being developed right now by thinkgeek after their April Fool's marketing joke backfired and they got swamped with calls. It'll have your choice of famous tracks based on mood. Amazing, huh? Pretty soon, it won't be ringtones, it'll be t-shirt tones.

[ggalen]

To write a song-composing program you have two basic problems to solve:1) Generating melodies and chordal harmonies to evaluate, and2) Correctly evaluating what's GOOD and is likely to be a hit...#2 is the harder problem. As we all know!Where are you going to get the rules about what's good?Remember, you want to recognize something that has not been done before, but is as GOOD or better than something that has been done before.I mean, let's say your pattern generator comes up with basically a cover of "Satisfaction" by the Rolling Stones.You have to recognize that even though it is GOOD and catchy...it is a BAD idea as a new song. Because it is not new.And these are the kinds of issues you'll run into.

[gongchime]

Hey matto, if there is a rule that you'll use the mixolydian scale, for example, such a rule is too rigid. How to account for chromatic notes which mixolydian music also utilizes. So, now we need a rule for when and where to use the chromatic pitches. We could switch to using a database of known jazz, blues or rock riffs, for example, but then all we would be doing is working through every combination of known riffs instead of coming up with new ones. If we say we could combine a rule based approach with a database approach then it starts to get really complicated. Performing transformations on the known database would work to some extent.But you're right to think what is it the working composers are actually doing? Explaining it is very difficult even for the people doing it. But still they take students and try to teach it. The sheer folly of trying to teach something as stupid as a computer, but still we try. I've seen the rules these expert systems are using after attempting to figure out what humans are doing. They're not complete enough yet, especially rhythically. Anyway, I don't want a computer that composes music. I want a computer that only HELPS ME compose music. It can do that by getting as close as a computer is capable of getting to what humans are doing. What are humans doing? We listen to a lot of music, then we try to create music similar to but not identical to what we heard. How is that accomplished? The answer is creative transformations of the original elements. Humans don't really apply the rules rigidly though. We might intuitively come up with an upward sequence in inversion but then tweak the last note because we didn't like it. The computer can't do this last step. The computer is not good at evaluating what is good. But we are. We might only apply the transformation after the downbeat or after the high note or just at the place we feel the music we wrote is not good. The computer doesn't really know the music was bad there and therefore has no reason to look for a solution only at that location so it either applies the rule globally or not at all. As for a computer program that goes through what you called a tedious and overly simplistic TDL list, I want software that does that!!! And it's not overly simplistic. It would be a good starting point for any composer, beginning or advanced, to come up with material faster. I think they would like it once they got it in their hands. Those transformations are musically relevant. They've all been used by working composers before, just not on that scale. Usually applied only to the level of the motif or phrase. So, you scan in the latest music, the computer starts to play back every transformation and maybe you only hear one thing you like. (You'll probably hear a lot more). You now have SOMETHING to start with, whereas perhaps prior to that, you had next to nothing. This process does not work so well for SONGS. It's more effective for instrumental music. Words have their own rhythms and you kind of have to stick with those. Starting with the music and then putting words is probably backwards for a lot of SONGWRITERS. So, I can see how this doesn't really work so well for them.Combining the melodic rhythm from one tune with the melodic series of another tune is a kick-ass idea and not simplistic at all. (Even if it doesn't work 75% of the time.)Also, individually, each transformation is simplistic and I had originally rejected them, thinking they only had value to dodecaphonists. But simultaneously applying the retrograde with the transposition and rhythmic displacement so that the next section of music does not start on the same rhythmic location or on the same note and now has a different melodic countour and melodic rhythm, consistently works very well for me in my opinion. It's what allows me to take a short Asian folk melody and extend it into a complete composition. Even if you don't like my CD's chords or lack thereof, or my instrumentation or my textures, MELODICALLY, I feel this method is quite a success at whatever level of complexity it is felt to have.

[milfus]

true, but #2 is in the hands of the user, itd be like next next next next, thats a keeper

[ggalen]

milfus,Except that since the computer is so poor at evaluating what a human would "feel" is good, it'll throw so many bad combinations at you that it would soon become tedious.More like: "Next...next...next...next...next...next...next...next...next...next...next...next...next...next...next...next...next... next...next...next...next...next...next...next...next...next...next...next...next...next...next...next...next...next...ne xt...next...next...next...next...next...next...next...next...next...next...next...next...next...next...next...next...next ...next...next...next...next...next...next...Well, MAYBE....next...next...next...next...next...next...next...next...next...next...next...next...When's LUNCH??! "

[gongchime]

Galen thats the reason I'd like the computer to be limited to tried and true transformations of existing material than attempt to make it come up with new material using rules, or combining pre-existing elements in a database or using fractals, etc... The standard transformations of transposition, inversion and rhythmic displacement are KNOWN to work enough of the time for it to be worthwhile to check them out. I don't need a computer for that but, Santa, I'd LIKE one under my tree if I can get one.Next, I'd like to talk about pattern languages for documenting fields of expertise.

[gongchime]

The first questions when creating a pattern language for music composition are what problems do you commonly face as a composer and what are the key characteristics of effective solutions?One question is where do you come up with melodic series'? Unsurprisingly I have several tedious lists. One for jazz, one for diatonic music, etc...the other name for them besides tedious list (TL) is knowledge database. Here is one possible list of common diatonic series used by me when playing rebab (bowed instrument the size of a viola held like a cello made of a coconut) and diatonic marimba improvisations which I suspect may also apply to pop melody construction. We all realize that this is only one answer to the problem. There are several other ways to approach it.I read that professional performers minimally are able to play for one hour without repeating themselves. This just scratches the surface in that case.12131443424113511112222333333332222111111122233333322211111223333221134231223212123453444444524514513571243431254354254112312412521324312323212154254354123123123232123212321212321232123121232343531342121341452453455214213213213213213432321212132142152175311234321234321432123432123432123412341234123443214321432112312312332132132123212212322123234345124123124131211424342135124645316531213571246843432321212344321545354535253525155125352535453545111211131141114443444244414111311121111444244434441112111311121113111411131114111551114111311141113111211131112111555455535554555355525553555255511555255535552555355545553555455512345654321234567898765456789 ten eleven ten 987623214565478987123423456756789 ten 896789 ten eleven twelve eleven ten 98765456789876654321234565432198987456541232198 ten 9876576543243218987565442321123245657898123423334534566543543243211242353468757646531212121221212121etc...Is access to this particular database an effective solution for a place to acquire the melodic series you're going to use for your next tune? I feel this database is insufficient by itself for consistently coming up with new melodic series but it is one place to turn for a couple of tunes and therefore not without value.The difference between a computer coming to the task of music composition and a person is that a person has had a lifetime of musical experience even if only second hand as a listener that they bring to the table. A computer is as dumb as a tree stump.All the music you've ever been exposed to is the knowledge database from which living people usually derive their melodic series' from.So, now we have to answer the question, what characteristics does the effective solution or solutions have to the problem of fitting a melodic series onto a melodic rhythm?Any takers?

[gongchime]

Since my professor played with Charlie Parker in Birdland, I consider him an expert. He thinks your computer (your brain) minimally needs 100 jazz cds for you to be able to call yourself a jazz musician. Now perhaps there are only between one and three tunes on each cd that really floats our boat so probably we are talking about 200 tunes to encompass the minimum database for composing in a style. The actual number of tunes for pop or rock may be smaller or larger.

[ggalen]

Gongchime,There is simply an immense amount of processing the visual cortex does for your eyes to recognize the scene in front of you.Yet everyone does this effortlessly. But if you wanted to write a computer program to do scene analysis, it is very involved and computationally a heavy load.I think music and the brain are similar. I think a similar amount of "processing" comes into play with music, and it is just too hard to "do it by hand", so to speak, by trying to come up with the processes that are done in our neurons and then programming a computer to do it.Look at the complexity of your most recent post. As you say, it just scratches the surface.For me, I am not able to think of music that way.I have to just let my "musical cortex" do it's wonders!Have you ever played with Band In A Box?It's a great "idea generator" for me.