Classification of Music Genres based on Mel Frequency Cepstrum Coefficients using Deep Learning Models

Genre classification is indeed a vital task today since the number of songs produced on a regular basis keeps increasing. On average, around, 60,000 tracks are being uploaded per day on Spotify. So, classifying these tracks by genre is definitely an important task for every  musical streaming services and platforms. Due to the high classification performance of neural network models such as convolutional neural network (CNN), multi-layer perceptron (MLP), and long short-term memory network (LSTM) are used in this work to automatically classify music into to its genres based on Mel-frequency cepstrum coefficients (MFCCs) instead of manually entering the genre. We experimented the models with the GTZAN dataset and provided a comparative analysis on the classification efficiency of deep learning models. We achieved a classification of 70.42% for our proposed CNN model which is greater than the human accuracy and over other deep learning models.