< Back to IRCAM Forum

Mubu.process chop overlapping segments

MuBu
#1

Hello all,

I am working on a project using corpus-based synthesis, which architecture is close to cataRT. I am currently using the chop algorithm for the segmentation of samples, and I wanted to know if there is a way to segment the samples with a sweeping process / by overlapping them ?
For example, if I choose a chop.size of 100ms and my sample is 1000s long, I wish that the algorithm could be able to analyse 100 segments instead of 10 if I can choose an “overlapping factor” of 10.
The algorithm would analyse these segments : 0-100ms, 10ms-110ms, 20ms-120ms, … instead of 0-100ms, 100-200ms, …

Thanks !
Ryo

#2

Hi Ryo, chop by default can not overlap segments,
but you can achieve this using a very large window (which becomes the segment) and not use chop.
Another workaround would be to analyse each sound 10 times with varying chop.offset.
HTH

#3

Thank you for your fast reply !
As you mention, I also thought that the several analysis iterations and varying the chop.offset would be a solution, but I am not sure how to practically do this.
Let’s say I have 100 samples to analyse 10 times.

  1. If I readappend all my samples 10 times in the same track, is it possible to process them differently according to their buffer position ? (0 to 99 with offset 0ms, then 100 to 199 with offset 10ms, and so on…)
  2. As I am using a mubu.knn object to later select the segment based on their descriptor analysis values (as cataRT), I am not sure if I can separate the analysis iterations across several tracks (and processing them separately).
    (btw is there a way to readappend folders ?)
#4

you need to script this, set bufferindex to the new buffer, set offset, then start process and loop

this is easy by marking the offset in the bufferinfo, or add :const to mubu.process to access it as a descriptor

readfolder