c:/nigel/breathy/labnotes.txt

#=== Procedure ===
matlab
addpath midlevel/src
cd breathy

findDimensions('social.tl', 'pbook_cpps.fss');
#on my work computer (32GB) takes about 4 minutes for each channel of each track

diagramDimensions('rotationspec.mat', 'pbook_cpss.fss');

applynormrot('social', 'pbook_cpps.fss', 'extremesFeb2022');


#=== History ===


The original work was done December 2020, and saved in the Dec2020 subdirectory.

Revised February 2022; results saved in the Feb2022 subdirectory.

Substantitive differences: 
- cpps is now track-normalized
- cpps is now computed over windows, rather than sampled at window centerpoints.
In essence, these fix some bugs of my earlier implementation.

Results of the change:
- the patterns discovered do not change at all, as seen by looking at the loadings
  figures for Dimensions 5 and 10, old and new.
- the extremes found change slightly, but for dim05 for every speaker the
  top one or two exemplars, for both high and low, are the same as before

Thus,
- There is no need to revise any conclusions of the paper.
- some of the stimuli selected were  not the very best exemplars,
  but the hypotheses were supported anyway, so there is no need to
  redo the experiment either.