A New Approach For Musical Genre Classification Utilizing Cnn-Lstm With Muti-Feature Integration

Abstract

This thesis investigates the task of music genre classification using a deep learning approach, combining the strength of the Visual Geometry Group Neural Network (VGGNet) and Long Short Term Memory (LSTM). This study aims to develop an effective model that can accurately detect genres on complex and large datasets, a challenging problem in music genre classification. The dataset used in this study consists of over 10,000 tracks, covering a wide range of genres. To handle the complexity and diversity of the dataset, the proposed model uses a parallel architecture that can extract features at different levels of abstraction. The model also comprises a convolutional branch for extracting high-level spectral features, a recurrent branch for modeling temporal dependencies in the audio signal, and a hybrid branch that combines all features. Overall, this study contributes to the field of music genre classification by providing a robust and accurate model for music genre classification, which can be applied to various real-world applications, such as music recommendation systems and content-based music retrieval.