Music Genre Classification Using Acoustic Features and Autoencoders

Music recommendation and classification systems are an area of interest of digital signal processing and digital music processing. In this study by using digital signal processing techniques and autoencoders, music features are extracted and then with these features music classification and clustering has been done, and with the results music recommendation has been made. Obtained results are compared with each other. In the study, GTZAN dataset has been used. Purpose of this study is to compare the result feature extraction with auto encoders and digital signal processing techniques. For digital signal processing, used methods are as following: Mel Frequency Cepstral Coefficients (MFCC) and it's derivative, Zero Crossing Rate, Spectral Bandwidth, Spectral Rolloff, Spectral Centroid, Spectral Contrast, Spectral Flatness, RMS (Root Mean Square Energy), poly features, Chroma CENS, Chroma CQT, Chroma STFT, tonnetz, Wavelet etc. For the classification part MLP Classifier, Logistic Regression, Random Forest Classifier, Linear Discriminant Analysis, K-Neighbors Classifier, SVM, Naive Bayes, Gradient Boosting Classifier, Ada Boost Classifier used for classifying the data.