App/territory has two modes of operation.
territory' mode, sounds are discretely arranged as if internally-homogeneous, while thecontinuous' view computes a position 20 times per second.
- Chromatic comparison
- Consider 88-pitch binary threshold features (ie. to link third harmonic with root)
- Number-letter codes
- Volume (and shape) of synthesis
- Smooth interpolation / non-points
- Output a human-readable score (?)
- Prune segmentation on duration (>100ms)
How to run
- run segmenter.py on long audio file(s) to break into units
- train pca-space with pca_freqs.py (on the segmenter output)
- run compute_coeffs.py to compute coefficients and render to json/js
- will save app-data-0.js
- will also dump pca & eigenvectors to stdout (copy them into dev.js (!))
How to make a new map based on relations established through use?
run pca_freqs.py to to generate a bundle.
put it in www/continuous and edit index/manifest
git clone git://git.numm.org/app-territory