An Improved Music genre classification using Convolutional Neural Network and Spectrograms

The entertainment industry and playlist generators like gaana, spotify, Slacker, etc. have become increasingly interested in types of music in the past few decades. In this study, Information Extraction (IE) and information methods of processing are integrated. In a broader sense, we humans undertake a lot of the work of classifying musical styles into their respective genres. Therefore, this article proposes a system for categorising musical genres using spectrograms. The spectrogram represents the results of applying the Short-Time Fourier Transform (STFT) to the sound files. It creates visuals by primarily focusing on frequency and timing of an audio source. The primary focus of this proposed research is on decoding acoustic signal into spectrum analyzer and using those as input to a classification. Convolutional neural networks (CNNs) are used as the basis for the classification system in this study. The GTZAN dataset was chosen for this study because it contains accurate classifications of ten different types of music (blues, classical, rock, etc.). Spectrograms are created, data sets are separated, images are resized, the model is tested with convolutional neural networks, and finally, the trained model predicts the genre based on its output. The recommended model will be tested and evaluated for its loss and accuracy rate in the end. Increased accuracy in identifying musical genres via spectrogram analysis of audio recordings is reported.