Using Million Song Dataset, an artist classifier has been developed. In order to this, different network architectures have been designed.
The scripts to train and get the accuracy of each network are in 01-Development_cod folder. In order to run this scripts is necessary to have a GPU.
There are 4 different network designs:
- CV_CNN_1_.py
- CV_CNN_2_.py
- CV_CNN_tesis_.py
- CV_CNN_articulo_.py
Each network has 5 different inputs to be compared:
- Xy_7_MFCC1.pickle --> MFCCs extracted with MFCC_1.py on songs fragments of 7 seconds
- Xy_7_MFCC2.pickle --> MFCCs extracted with MFCC_2.py on songs fragments of 7 seconds
- Xy_14_MFCC1.pickle --> MFCCs extracted with MFCC_1.py on songs fragments of 14 seconds
- Xy_14_MFCC2.pickle --> MFCCs extracted with MFCC_2.py on songs fragments of 14 seconds
- Dataset_MFCC_h5_24-48_segs.pickle --> MFCCs precalculated by the dataset developers on songs fragments of 24-48 seconds
- Dataset_MFCC_h5_12-24_segs.pickle --> MFCCs precalculated by the dataset developers on songs fragments of 12-24 seconds
- Dataset_MFCC_h5_6-12_segs.pickle --> MFCCs precalculated by the dataset developers on songs fragments of 6-12 seconds
These inputs are not available in this repository. Nevertheless, if you want to have them in order to run the scripts you can contact me via email (flavi13a@gmail.com).