Henning Thielemann
On Thu, 5 Jun 2008, Achim Schneider wrote:
The recent discussion about Markoff chains inspired me to try to train one with all the Bach midi's I have on my disk, collecting statistics on what intervals tend to get played simultaneously, which follow others and in which way the pitch offsets from its mean, so that melodies fall and raise "naturally".
I don't know, if you already found that one: http://darcs.haskell.org/haskore/src/Haskore/Example/Kantate147.hs
Surprisingly I also tried Markov Chain on a Bach song. But my approach was too simplistic in order to produce a nice new song.
Yes, you need to take both dimension of music into account, that is time and polyphony. Bach uses quite exceptional polyphony from time to time, but it always stays harmonious: You have to have eg. a 0% probability of ever playing a note and its minor second. The probability of a note and its quint will most likely be at least 50%, but then there are chords that sound atrocious if it's there. What I need is basically one view of the data as list of used chords, and one graph of all possible time-linear progressions... that is, voices, for a definition of "voice" that makes the guitarist in me shudder. Seems like I'm going to make close acquaintance with fgl, after all. -- (c) this sig last receiving data processing entity. Inspect headers for past copyright information. All rights reserved. Unauthorised copying, hiring, renting, public performance and/or broadcasting of this signature prohibited.