Search code examples
generatorhidden-markov-modelspuredata

Pure Data for music composition?


I am currently working on a project where a Classical Music Generator is set to be built. A lecturer who is pretty much a expert in sound processing suggested that I could construct a Hidden Markov Model which will generate and return compositions.

There are already 80+ MIDI files from various renowned composers (Mozart, Bach, Beethoven...) in my computer ready for training. I just wonder will Pure Data be a suitable language for constructing HHM(s) which will finally return a generated music piece, from some MIDI inputs?

And if yes, any advice on how to utilise those MIDIs and put them into pieces for the HHM?

Thanks in advance.


Solution

  • Take a look at this library. It has HMM and a number of other machine learning algorithms that you could use.

    https://github.com/cmuartfab/ml-lib

    You can also use HMM for signal related tasks. There is a paper here that you may find informative:

    http://tcts.fpms.ac.be/publications/papers/2013/dafx2013_mage.pdf