Takes a collection of audio files (sources) and a single longer audio file (target) and slices them all up. Then it replaces each slice in the target with the most similar slice from the sources, where "most similar" is determined by feature extraction and a 1-nearest-neighbor search.
Run from slicer.py. Edit variables at the top of the file. Change stuff around in feature_extraction.py if you want to change what features are used, but be careful to make sure that every signal input ends up with a feature vector of the same size.
N.B. The source should be a path to a folder, and all sound files in that folder (and all subdirectories) will be included in the sources. If a source is sufficiently long (8 seconds by default, editable by parameter) it will be sliced up, but otherwise it will be considered a source slice on its own. The path to the target should go directly to one file.
Requires scikit-learn (for PCA and k-nearest-neighbors) and librosa (for onset detection / feature extraction).