3 subprojects:
- ZRTools: original repo (with a few modifs). Contains the c code and scripts
- plebdisc_interface: python interface for ZRTools functions. Extension of a few functions
- plebdisc_tools: python scripts (wrapper for main function and several analysis tools)
Unsupervised spoken term discovery.
Split plebdisc in 4 subfunction and expose those to python. Numpy interface with those subfunctions.
- compute_similarity_matrix
- filter_matrix
- find_matches
- refine_matches
- main: - call lsh.main, plebdisc.main (h5features interface) - estimate "good" parameters
- io: load binary files in numpy arrays + a few operations on those.
- evaluate: - final step (ned) WIP - matches (precision, minimal achievable ned)