Ah - found your question.
First and foremost I am not sure computer generated music is the way to go - it would be far easier to get a musician to record some MIDI files for you and then you can play them back in different ways - presumably there are plenty of musicians in a university...
failing that I would expect you will have to make different models for the different types of music. The Markov Chain approach will work for some of them but it misses out huge chunks of what different music types are about - for example punk is musically very simple but what stops if being dull as dishwater is the anger and energy put into it by the performers - computer generated punk is probably a lot harder than generating something like Bach or Phillip Glass style music which is more about the structure and composition. Like wise the blues is very simple musically but is about the experience of the musician - I would suggest that computer generated blues would be very close in experience to "Richard Clayderman plays..." (elevator music of the lowest kind)
cheers
mark
-----Original Message----- From: pd-list-admin@iem.kug.ac.at [mailto:pd-list-admin@iem.kug.ac.at] On Behalf Of Nicolau Werneck Sent: 02 September 2002 18:37 To: pd-list@iem.kug.ac.at Subject: [PD] [OT] Sensitivity to timbre and adequacy of instruments to musics
Hello, people.
Sorry if I'm off-topic, but I think maybe I could reach people here who would be interested on this.
I'm beginning a research at my university, and I'm looking for comments and suggestions from other people.
The research goal is to catalog correlations between kinds of music (that is, musics with similar structures) and the timbre of instruments that are selected by people to play them. Basically, we are going to play computer generated music to listeners, and ask them if the instrument used is adequate to the music. We expect to find, for example, that surf music is better with a lot of reverberation, punk rock is better with a lot of overdrive, blues is better with just a bit of overdrive, and so on...
The project will be developed in two parts: first we will measure the sensitivity that people have to variations on the parameters of the models of the effects. Then we will play the musics, and find the correlations.
It looks like the worst problem we will find is on the composition of the musics. I'm thinking about finding stochastic models to various styles of music, and than program a music generator. Does anybody here have any experience on generating random music in PD? What would be the best approach? Making an external program that export the song in MIDI, and bring it to PD?
...Thank you, bye!!
On Tue, Sep 03, 2002 at 03:44:01PM +0100, mark wrote: Hello. Thank you all for answering!
First and foremost I am not sure computer generated music is the way to go - it would be far easier to get a musician to record some MIDI files for you and then you can play them back in different ways - presumably there are plenty of musicians in a university...
Sure!... Playing "real" musics would be probably easier to implement. But since I want to play different musics that share some structural characteristics, we would have the problem of analysing the musics and putting them together. Generatting the musics I'll be *shure* that they have the same structural qualities, because they were made this way! So we change the natural and difficult-to-analyse musics by unnatural musics that certainly have the same characteristics. This way we avoid the problem of having to define musical styles, and jump right to the analytical domain...
The Markov Chain approach will work for some of them but it misses out huge chunks of what different music types are about - for example punk is musically very simple but what
Oh, I love Markov chains... In fact, I started this work because I saw they would show up some time :D
I agree with you that punk and blues will sound awful... A good part of blues compositions can be fitted under 3 or 4 basic structures, and playing around with them won't sound nice. But that doesn mean that I'm not curious to try it it and find it actually bad :) .
This problem suggests that implementations for this kind of music would require an dictionary approach, while Bach would be more algorythmic...
I've seen jazz improvisation softwares also, and it looks like they just pick notes at random inside scales related to pre-programmed chord progressions... I don't know if it sound good :/
I'll make some experiments by next month, and I can post the results here!
bye...
<<Nicolau;
Hi, Nicolau Werneck hat gesagt: // Nicolau Werneck wrote:
I've seen jazz improvisation softwares also, and it looks like they just pick notes at random inside scales related to pre-programmed chord progressions... I don't know if it sound good :/
If you speak of "Band In A Box", I think, their model of generating improvisations or the band is much more complex. BIAB can actually improvise in different styles and make a sax player really sound like Charlie Parker for example. Not regarding intonation of course, because BIAB only generates midi, but the lines that are played, and the typical Parker-licks do actually resemble Parker soli. So a lot of analysis went into BIAB, but unfourtunatly it seems to be a company secret, how they did it.
Frank Barknecht _ ______footils.org__