Automatic music classification for humans
-e <feature> calculate feature, rewrite feature file
-u update base (add new songs)
-r rebuild base (recalculate all features)
-t calculate features for reference playlist (from music/tags/)
-t <tag> most relevant bands and songs for choosed tag
-p <tag> generate .m3u file of top songs for choosed tag; playlist saved in data/playlists/
-a <artist> tag relevancy values for artist
-f <tag> most important features for choosed tag
-s <tag1><sign><tag2> most relevant bands and songs
for two tags combination; supported signs: "+", "-"
-d <artist> most important features for choosed artist
classical, dance, electronic, happy, melancholia, rock, sad, trash
- bpm
- autocorrelation
- zero-crossing
- onset_strength
- onset_regular
- centroid
- volume
- spectral flux
- spectral contrast
- spectral std
- spectral median
- spectral bandwidth
- spectral rolloff
- cepstral coefficients (n=10)
Classifiers trained with reference playlist data (data/features/tags/)
Default path for your songs: music/favourite/. Repository has precalculated features for reference and sample songs.
Classification algorithm is gradient boosting. Feature selection algorithm is recursive feature elimination (RFE).