Is there a way to analyse material in CataRT so that the segments can be described by the amount of transient or tonal structure in the sounds?
Great question! My first idea would be:
Tonality
- should show in mean periodicity
Transients
- might show in standard deviation of loudness and/or periodicity,
- or you could run a secondary onset detection and count the onsets/s of your larger segments
Also look at the noisiness/harmonics related features of ircamdescriptor.
In any case, it will help to constitute some examples of each content, in several levels (e.g. no/mid/high transient content) and combinations.
HTH
Thanks Diemo. Creating specific example material to try and understand the different descriptors is a great suggestion. Loading in pure noise and sine tones alongside short pulses at different frequencies definitely reveals a lot about how the system works.
Periodicity seems to separate transient and tonal sounds but it depends on the texture and shape of the segments and the analysis method.