Music is categorized into subjective categories called genres. With the growth of the Internet and multimedia systems applications that deal with the musical databases gained importance and demand for Music Information Retrieval (MIR) applications increased. Musical genres have no strict definitions and boundaries as they arise through a complex interaction between marketing, historical, and cultural factors.
Our project is a research – based analysis of the music genres using spectral and rhythmic features of the music We have done a comparative study using various machine learning algorithms to classify the music into its various genres namely, blues, classical, country, disco, hip-hop, jazz, metal and pop respectively. We have used various audio features, such as Mel Frequency Cepstral Coefficients (MFCC), Delta, Delta- Delta and temporal features, including beats and tempo to featurize our data. Various classification algorithms, such as Support Vector Machine(SVM), Decision Tree, k-Nearest Neighbors(KNN), Random Forest and Gradient Boosting Classifier are used in the classification of the data.
The dataset we have used for our music genre classification is GTZAN[]. This dataset contains 1000 song files, each of which is 30 seconds long. These songs are classified into 8 genres, namely, blues, classical, country, disco, hip-hop, jazz, metal and pop respectively. The sampling rate we have used for our data files is 22050 Hz. All these files were in .au format which were converted to .wav using online converter. This is done since .wav format files are easily read by python modules. We divided our dataset into training and testing data in the ratio 7.5 : 2.5. Spectogram of songs of different genres are depicted. The dataset is available at this link : marsyasweb.appspot.com/download/data_sets/
We have uploaded a small video explanation of our machine learning code as well as the flask API along with the research papers referred and the research paper we are still working on this link - https://drive.google.com/open?id=1oAtUAaVEIwL_rsgMMQ_lYtfMLpl-oPXr