An attempt to classify genres across music and imitate behaviour of popular streaming services
#Dowloading_datasets We use GTZAN genre collection dataset for classification. http://marsyas.info/downloads/datasets.html This dataset was used for the well known paper in genre classification " Musical genre classification of audio signals " by G. Tzanetakis and P. Cook in IEEE Transactions on Audio and Speech Processing 2002. The dataset consists of 10 genres i.e Blues Classical Country Disco Hiphop Jazz Metal Pop Reggae Rock Each genre contains 100 songs. Total dataset: 1000 songs
All the .au files were converted to .wav using Pydub library or can also be done in the command line directly by changing extensions
'tempo','zcr','spec_bandwidth','rms','spec_centroid','spec_rolloff','chrom_spec' Data has extracted using Librosa library. Features Extracted were the following:-
1)Zero Crossing Rate
2)Spectral Centroid
3)Spectral Rolloff
4)Mel-Frequency Cepstral Coefficients(mfcc)
5)Chroma Frequencies
6)Spectral Bandwidth
7)RootMeanSquare
8)tempo
And these 8 features appended to give a 57 length feature which is later reduced to 37 length feature
Used to identify the redundant features
Diffrent classification and Deep Learning models were used to classify data. Diffrent parameters were used to tune the model
SVM gave the highest test accuracy with .82 percent and seemed to perform very well in predicting Classical, Metal,Pop genres. But gave poor accuracy with Reggae, Disco, Country genres