Hi, i'm looking for a way to make a patch that recognizes musical motifs from a flute player in real-time.
Maybe using ann externals to first training with the player, or with some recorded material.
There is other ways to do that?
Someone is doing something similar on Pd?
Pd is the best choice to do that work?
Sorry for many questions. Thanks for atention.
cris
Hi, i'm looking for a way to make a patch that recognizes musical motifs from a flute player in real-time.
The first problem is how accurate you expect the pattern recognition to be. I have a piece where I ask Pd to make binary descriminations on input from the flute player, in real time. It works fine. It is simple enough to get note information from fiddle~ and write that as a list or a table. The problem is that you won't get 100% of the flute notes (legato notes sometime don't trigger etc.) and the notes fiddle estimates are not always perfect. The next problem is how fast can pd process the list/table to identify pattern. That will depend on the length of the patterns (of course) but also the narrower the definition of what pd needs to look for the faster it can match (probably). To create a system that works for you, you will definately need to start with recordings before you have something to work with a player on.
There is a fair ammount of research on music pattern recognition out there, though real-time is not a priority for most at this stage.
Oded
the best performance I ever heard using musical pattern recognition was by the trombone player george lewis http://en.wikipedia.org/wiki/George_Lewis_(trombonist) he uses a max patch called voyager. I have no idea how it works, maybe you find some documentation online. he worked on it for several years. and it was extremely good. marius.
Oded Ben-Tal wrote:
Hi, i'm looking for a way to make a patch that recognizes musical motifs from a flute player in real-time.
The first problem is how accurate you expect the pattern recognition to be. I have a piece where I ask Pd to make binary descriminations on input from the flute player, in real time. It works fine. It is simple enough to get note information from fiddle~ and write that as a list or a table. The problem is that you won't get 100% of the flute notes (legato notes sometime don't trigger etc.) and the notes fiddle estimates are not always perfect. The next problem is how fast can pd process the list/table to identify pattern. That will depend on the length of the patterns (of course) but also the narrower the definition of what pd needs to look for the faster it can match (probably). To create a system that works for you, you will definately need to start with recordings before you have something to work with a player on.
There is a fair ammount of research on music pattern recognition out there, though real-time is not a priority for most at this stage.
Oded
PD-list@iem.at mailing list UNSUBSCRIBE and account-management -> http://lists.puredata.info/listinfo/pd-list
cristiano listas wrote:
Hi, i'm looking for a way to make a patch that recognizes musical motifs from a flute player in real-time.
Maybe using ann externals to first training with the player, or with some recorded material.
There is other ways to do that?
Cris,
While recent research has demonstrated solid progress, this is a _very_ deep subject with many remaining difficulties. The primary issues are best demonstrated with a few questions:
1 - What specifically constitutes pattern similarity during the matching process? Is it a rhythmic pattern? Pitch? Are both required for identification?
2 - How much (and in what ways) can the input deviate from the expected pattern and still be recognizable?
3 - What type of "extra" events are allowable without damaging the integrity of the original? For example, if 5 consecutive "extra notes" were detected at the start of your motive would you want the system to identify it as the original, a variation, or might it be part of a new/different (but possibly related) idea?
Here are a few highly accomplished and respected authors on the subject. Their research is a quick google away...
Emilios Cambouropoulos Roger Dannenberg Pierre-Yves Rolland David Temperley
HTH, Greg
Hallo, Dr. Greg Wilder hat gesagt: // Dr. Greg Wilder wrote:
Here are a few highly accomplished and respected authors on the subject. Their research is a quick google away...
Emilios Cambouropoulos Roger Dannenberg Pierre-Yves Rolland David Temperley
Also Robert Rowe's "Machine Musicianship" is a nice intro to this and related topics.
Ciao
Frank Barknecht _ ______footils.org__
-
cristiano listas -
Dr. Greg Wilder -
Frank Barknecht -
marius schebella -
Oded Ben-Tal