An Enhanced Automatic Song Genres Classification Using Deep Learning Technique

  • Mr.A. Sudhakar, V. Bhargav Durga Prasad, G. Krishna Reddy, T. Sai Nikhil, B. Komal Sai


As humans, we can easily classify two songs by listening to them, but in the case of automating the classification, Machine Learning comes into the picture. Classical ML algorithms can make a decent classification based upon temporal (Zero Crossing Rate, signal, etc.) and spectral (centroid, roll-off, flux) features that are dependent on time domain and frequency-based features. By understanding music on a deeper scientific level, this project aims at classifying various songs into genres. We used a neural network model in this project, i.e. a CNN based on [1]GTZAN dataset consisting of 1000 songs in the format ".wav." Among 1000 songs each genre consists of 100 songs and each song is of 30 sec time period. We mainly focused on Mel-Spectrogram to extract features from a given song as well as while designing the model. In the case of using Mel Spectrograms CNN model outperformed all the classical Machine Learning models results.

How to Cite
B. Komal Sai, M. S. V. B. D. P. G. K. R. T. S. N. (2020). An Enhanced Automatic Song Genres Classification Using Deep Learning Technique. International Journal of Advanced Science and Technology, 29(3), 5899 - 5903. Retrieved from