Music style classification by jointly using CNN and Transformer
Rui Tang, Miao Qi, Nanqing Wang · 2024
Music influences people in many ways and plays an important role in human life from emotional expression to social interaction to cognitive development. However, the variety of musical styles is often difficult to distinguish. In this paper, different from existing methods that music presented in the form of audio information can be classified as a sequence of features divided by time through RNN or LSTM, a novel music style classification method is proposed by transforming music audio into audio image. Moreover, Convolutional Neural Network (CNN) and Transformer are combined to jointly extract rich audio image features for music style classification. The effectiveness of the proposed method is evaluated by a large number of ablation and comparative experiments. The experimental results demonstrate that the classification accuracy of our proposed method can achieve satisfactory classification accuracy and is better than some existing classification methods on GTZAN dataset.