Utilizing Spectrograms and Deep Learning Techniques for Improved Music Genre Classification

In the modern world, music occupies a major and possibly even vital position in our day-to-day behaviors. The variety of musical genres is extraordinarily extensive, and each genre possesses its own set of traits that are unique to itself. Individuals are able to acquire their own distinct musical preferences and styles as a result of this variation. Accordingly, it has become a relevant and contemporary problem to accurately categorize music as well as propose new tracks to consumers on music streaming services and purposes. This is because of the fact that music is always evolving. In order to effectively solve this difficulty, one of the most successful techniques is to categorize music according to the genre it belongs to. This method not only helps in organizing the vast selection of music that is available, but it also improves the user experience by delivering customized suggestions based on the user's interests regarding the genres of music. In this investigation, we attempt to solve the problem of automatically categorizing different types of music by recasting it as a query using pattern recognition. Generation of spectrograms from the sound waves was the method that we utilized in order to approach the content analysis of musical compositions through the visual domain. Through the use of these spectrograms, which serve as graphical representations of the audio information, the subtle patterns that are present within the music are captured. We utilized a wide range of machine learning and deep learning methods in order to successfully recognize and categorize these patterns. For the purpose of representation learning, we specifically utilized convolutional neural networks (CNNs), support vector machines (SVMs), ResNet-50, and AlexNet. The selection of these algorithms was based on their capacity to effectively manage complicated data and to derive relevant characteristics from spectrograms. For the purpose of our research, we made use of the GTZAN database, which is a dataset that is internationally recognized in the discipline of music genre classification. With regard to the various algorithms that were utilized, AlexNet was able to attain the maximum level of accuracy in the classification of the music genres, so demonstrating its efficiency in this endeavor.