A couple of weeks ago, I wrote a blog explaining how I wrote an orchestral composition. While I was working on the actual piece, scientists from Imperial College London stated that music compositions can be created through a evolutionary Darwinian process (report from the BBC here). The research claims that you can start off with a simple combination of notes (or just some noise) and modify this based on feedback from listeners. The system learns how to create more music through a process of natural selection: music judged favourably 'survives' and makes into in the piece, whereas music judged unfavourably is removed. This process continues until a full piece of music is composed. Thus, the authors conclude, music can evolve naturally without ever involving a human composer.
It's true that a lot of music theory is very objective. Certain combinations of notes and chords work together well and others don't. Even with a basic grasp of music theory, it's possible to create a piece of music that sounds good without even hearing it. An obvious example of this is the work of deaf composer, Beethoven.
However, what's important to note about this research is that it assumes an entirely 'bottom-up' approach to music composition. That is, that composers start with a simple idea and keep changing it until it sounds right. Personally, I think this is quite a worrying view: I believe that good music composition (like any art form) is not produced by a bottom-up (trial-and-error) process; it's a top-down process: Good composers can envisage a finished piece of music before even transcribing a single note: they know how the composition sounds before they've even started it. The only alterations needed are when their working compositions are a departure from what they originally envisaged, so adjust their work until it fits their idea of a finished piece. The process of doing this may indeed be bottom up, though it is governed primarily by a larger top-down process.
These are only my thoughts - I'be be interested to hear what you think, so please comment below!
It's true that a lot of music theory is very objective. Certain combinations of notes and chords work together well and others don't. Even with a basic grasp of music theory, it's possible to create a piece of music that sounds good without even hearing it. An obvious example of this is the work of deaf composer, Beethoven.
However, what's important to note about this research is that it assumes an entirely 'bottom-up' approach to music composition. That is, that composers start with a simple idea and keep changing it until it sounds right. Personally, I think this is quite a worrying view: I believe that good music composition (like any art form) is not produced by a bottom-up (trial-and-error) process; it's a top-down process: Good composers can envisage a finished piece of music before even transcribing a single note: they know how the composition sounds before they've even started it. The only alterations needed are when their working compositions are a departure from what they originally envisaged, so adjust their work until it fits their idea of a finished piece. The process of doing this may indeed be bottom up, though it is governed primarily by a larger top-down process.
These are only my thoughts - I'be be interested to hear what you think, so please comment below!
Tis an interesting thought!
ReplyDeleteI reckon that in order to envisage a piece of music in the first place, a bottom-up process is used; a composer starts from somewhere, be it a single note, or a sound or small part familiar in another song, and builds on it using (implicitly) their prior knowledge of what notes and instruments they like to hear together.
Infact, it is just a difference in the order of events. A composer starts from nothing and builds on it, whereas the algorithm you described (probably a genetic algorithm) starts with a whole "composition" in mind, and continually reworks it until it is judged to be good enough.
I see no reason why a more advanced algorithm however could replicate the kind of music a composer would make, and do so using a similar process.
Good points!
ReplyDeleteIt's true that some processing is involved in order to get to this 'finished' composition. The question is: how do you get to that point in the first place? If any bottom-up processing is involved in the formation of this idea, it must be a purely implicit, timeless process. This algorithm also depends on a (human) third party to rate it throughout this process, whereas in reality a composer relies on their own intuition to form their finished composition.
I can't yet envisage an algorithm capable of doing that!