This is an on-going research project started in the early 70's using destabilized Markov processes. The research is primarily for my own self-abusement rather than any scientific endeavor. The latest effort has used artificial neural networks to compose music. Training data came from early Beatles music. Several different neural network architectures have been used: Kohonen, feed-forward with backpropagation training, and several hybrid networks using a combination of Kohonen, feedforward, and ART.

    Here's a couple. I'll put some more samples here as soon as I have time to clean them up a bit.

    These were composed with artificial neural networks using pre-Yellow Submarine Beatles' songs as training data. I went in and tweaked here and there but the overall composition is ANN. The ANN was set up to compose phrases of chords and interlinked melodies. They are all in MIDI format (so your browser needs to know how to play MIDI files). The MIDI card used is a Roland LAPC so your instrumentation may vary.

    If you use them as examples, please let me know. If you want to modify them, I may have some problems. Let's talk. (I do appreciate the comments from teachers who've used the music as examples.)

    Chinatown, USA is an example of two contrapuntal simple melodies. The song lasts 58 seconds and is 1649 bytes long.

    A Day at the Beach is a longer piece with a little more complexity. I feel the need to come clean about one thing: I did put in the rock-n-roll riff as a bridge near the end. The percussion was also done by ANN but is VERY simple. I put a section in the code for the special effects bridge. The song lasts 2:59 and is 7650 bytes long.