• Advertisement
Sign in to follow this  

Representation for music

This topic is 4438 days old which is more than the 365 day threshold we allow for new replies. Please post a new topic.

If you intended to correct an error in the post then please contact us.

Recommended Posts

Does anyone have any references or ideas for representations of music for pattern recognition? -Kirk

Share this post


Link to post
Share on other sites
Advertisement
Music? You mean like manuscript type or analog sounds? I would probably use <note frequency, duration, volume profile over the duration, instrument, k-discrete sequence index>


Hehe, sounds like a fun project brewing in your mind.

Share this post


Link to post
Share on other sites
I was at a conference in 2003 where someone was showing off a GA that could generate the instrumental part of a composition when given a melofy. It was a pretty cool concept.

Share this post


Link to post
Share on other sites
Not manuscript, but rather analog. Also, rather than just sounds, I'd be interested in a representation for a whole song or performance. Most of the features you (Nick) mentioned appear to be associated with a single note as opposed to a (short) composition of notes.

Interesting GA application, too. That's something I'd like to see.

As for what's brewing in my mind, I have too many ideas. 8^) Make the voices stop!!

-Kirk

Share this post


Link to post
Share on other sites
hehe.. the voices are crazy, especially the one of the rabbit with long ears that plays the violi... oh wait you don't get that one too?

What I meant was to group those notes in sequence and you could have several instruments playing at once. What? You can't read minds?

Anyway, I'm not sure what you are driving at on the analog... but you could extract those features from an analog signal I suppose by using an FFT or Wavelet transform... (I venture to guess here...)

Share this post


Link to post
Share on other sites
Wouldn't MIDI be a good way to represent the music? MIDI events are easily analysed and processed, and also come quite close to being able to reproduce the original, at least on harmonic, rhythmic, and melodic levels.

Share this post


Link to post
Share on other sites
Hmmm...interesting...MIDI...hadn't thought of that. I'll look into that one...

thanks!!

Share this post


Link to post
Share on other sites
I've come across the use of RBF networks (radial basis function networks) recently for classification of music, so you might get some ideas from reading that literature. Should be easy to find online (that's where I saw it, but don't remember the authors off-hand).

As for representation of sounds, you might consider computing the complex cepstrum and performing classification in the space of coefficients of this transform.

Cheers,

Timkin

Share this post


Link to post
Share on other sites
Timkin,

Thanks for the references to RBFs and the ideas for representations. I have a feeling I'm going to have to do a lot of reading. 8^) Is it any wonder we love this stuff??

-Kirk

Share this post


Link to post
Share on other sites
whoa big words! *falls over*

While I bookmark the searches for later reading, I'd suggest Midi too. I've seen analyzers that get input from a microphone and store the result on a Midi. You could do that, and then check for patterns in midi.

Of course, you'd lose details like vibrato and such. Which makes it simpler to look for patterns.

Share this post


Link to post
Share on other sites
MIDI is a good idea for another reason. While researching this stuff for an idea I had involving ant colony optimization and music composition a few years back, I stumbled across the fact that there are text-2-midi converters. So, you can specify the music sequence output it in text and then have it converted to midi automatically without messing with any binary.

So, if you're working on a note level, then midi is definitely the easy way. Also, just google text to midi conversion and you'll probably find it in C++ or C, I think.

Share this post


Link to post
Share on other sites
One of my proffessors from back when I was in college, David Cope made a program that would generate new songs in the same style as inputted ones via some kind of pattern recognition, I forget the details of how it worked. But it used MIDI format.
you can probably google for it, the program was called EMI (Experiments in Musical Intelligence)

Share this post


Link to post
Share on other sites
Sign in to follow this  

  • Advertisement