At the forefront of Artificial Intelligence

Home » Articles » General » Applications

Applications in Music

The Artificial Intelligence applications in music are endless - unfortunately, at present there is very little so show for it. Artificial Intelligence and music are at either ends of the spectrum, AI being seen as the epitomy of computer science, and music the epitomy of art and abstractness.

Dynamic and autonomous music creation has endless possibilities - pieces could be composed in seconds, for garden centre, elevator and trance music! Computers could provide brilliant 'jam' partners for guitarists, blues/jazz players to help develop their style. Computers could provide piano accompaniment for orchestral practices. The possibilities are endless! Yet how can all of this be achieved?

Generating Computer Music

Creating music takes a certain amount of inspiration to get a piece started. How do we inspire computers? As of yet, there is no music-generating program that takes no human input. Most programs at the moment require the human user to set various parameters and options - these are used by the computer to generate the piece.

A lot of music programs use genetic algorithms as a means of generating pieces. GenJam by Al Biles uses genetic algorithms to create jazz improvisation riffs. The program would come out with a riff, and Biles would tell the program whether it was good or bad, thus improving the fitness of the riff. After much training, GenJam is a formidable improvisation partner!

Other program use several modules communicationing with each other - some GA-controlled, others not controlled to create a piece. The non-GA controlled modules use mathematical formulas to form note patterns, chords etc. Nevertheless, the largest direction in computer-generated music seems to be genetically-created music.

Why is this? Well, GAs do seem well suited to the problem. They set out to search for the best note sequence in an infinite search space. Given initial criteria, they may find some relatively good sequences. It is the fitness function that is so hard to program, because exactly what does make a musical piece good? For some, it has to have structure, flow and definite movements - for myself, it has to evoke an emotion of sorts, or have an excellent guitar solo! Therefore, creating a fitness function for these kind of goals is incredibly difficult.

What's more, music has to bind together - it is no use getting an genetic algorithm to generate 20 bars of music, and expecting them to gel together. There is a huge likelihood they'll make no musical-sense. So, how do generate something on the fly, yet allowing it to bind together? Fractals spring to mind for me. Fractals are used to generate landscapes on the fly that tile perfectly, since they often use Fourier Transforms.

Fractal Music

What do fractals have to do with music? Ever since the 1920s with Joseph Schillinger, music has been recognized to have a chaotic and recursive nature. Many other studies as to why we find certain music pleasing, and other music as cacophony. It has been shown that music often has spectral density of 1/f (the concept of spectral density is unimportant here). It has been shown that most fractals fall into a similar 1/f category too. Fractals have been used to generate music in several ways - you can select a row and use each pixel to represent a certain note. Other ways have been done by creating music in the same way that the fractal is drawn, with each pixel position represent a certain note. For me, the best example of a fractal-generated piece has been one created from a Mandelbrot set:

mandel1.mp3 (347Kb)

This is a very chaotic piece, with occasional breaks - nevertheless, it has a certain Eastern tone to it. The is a lot of evidence that fractal music could soon provide us with some very real, very entertaining pieces in the near future. If you are interested in finding more about fractal music, please see our links section.

Automated Transcription

Automated Transcription would indeed revolutize the music industry - imagine putting in a CD, pressing a button and having the computer create a perfect score of the piece. Transcription can be incredibly difficult, especially for pieces like Canon music, or new highly-layered pieces such as the works of Frank Zappa or Steve Vai. The human ear (well, the human brain) has the ability to listen in on a certain sound, ignoring (or at least not taking as much attention to) the other sounds. For example, when you listen to a song you can listen to the words without being distracted by the music, because you can concentrate on the singer's voice. If you want, though, you can listen to the guitar, bass, or drums without any trouble.

Creating a program to be able to hone in on these sounds is inherently difficult, since we have no idea how the brain is able to distinguish sounds within sounds. The area of voice recognition may some day lead us to answers, since voice recognition is basically the study of finding meaning in one sound - auto transcription would be finding 'meaning' (individual instruments) in a multitude of sounds.

Other Applications

While these two are the main areas of use for Artificial Intelligence, being a guitar player, I see other areas. Roland released a software package a few years ago - a MIDI program for guitarists. It printed MIDI files in terms of piano keys, music, or showed the fret board and the positions played (essentially, tablature). Now, the one complaint was that the tablature feature was terrible, since the notes and positions the program suggested had no 'logical' order. For non-guitarists, on a guitar you can play a note in several areas, in fact on my guitar I can play a certain note in 6 different places (E). Therefore, when playing a piece, you can play a certain note sequence in different areas - these areas can make it easier or harder to play a piece, depending on the stretches, jumps and string skipping required. I have often contemplated creating a MIDI to TAB program, that takes a guitar riff and uses a genetic algorithm to find the best tablature to play it with.

Last Updated: 08/05/2000

Article Toolbar

BibTeX entry