Hello!
I’d like to import my own segment markers, and for mubu.process to analyse and segment the sound based on my imported time-chunks. I am able to import the marker times with mubu.track, but can’t work out how to tell mubu.process to do the rest of the job. Up to now I’ve been following the ‘howto-segmentation’ help patch, and I’m sure the answer it looking straight at me, but I can’t work it out. Super if you can give an example, or point me to the correct help patch.
Thanks!
Natasha
Hello,
I did a bit of work on segmentation, feature extraction per segment and thumbnailling for a musical AI. You could go with two directions:
1- Using a temporary mubu container: There is an example that comes with MuBu package that implements this. The particular subpatcher file is under patchers/mubu.stats.marker.track.maxpat. I can’t remember which main patch was using it though. Basicly, there is a temporary mubu container that records only the current segment, then applies feature extraction, and then follows that with statistics of features for thumbnailling.
2- Using a ordinary coll for markers: that is how I do it, just for the sake of using less memory. Given that your markers are time locations as opposed to frame numbers, you would need your mubu track to be time stamped. You can create a coll file with your markers, and ask the mubu.track to output particular indexes that is within your marker range. In my implementation, I just go through the whole mubu track, and whenever a marker is passed, I just save all feature statistics and clear all stat objects. This is a nastier implementation because all stats are calculated outside of mubu, which is a perfect recipe for a spaghetti patch.
That is being said, it seems that there is no easy bulit-in way to do it. If there is a simpler, fancier way; I am would love to use that.
I hope that helps!