An Attention Based CNN-LSTM Hybrid Approach for Music Genre Classification

This research designs and implementation of a music genre classification system that leverages an innovative hybrid approach combining Convolutional Neural Networks, Long Short-Term Memory networks, and an attention mechanism. This integrated system is adept at capturing the intricate time-frequency features and long-term dependencies inherent in music signals. The CNN component excels at extracting abstract features from audio's mel-frequency cepstral coefficients through a series of convolutional and pooling layers, while the LSTM network adeptly handles long-range dependencies within the sequential audio features. An attention layer is introduced to pinpoint and emphasize salient features, thereby enhancing the classification accuracy. The results from our experiments on the GTZAN dataset are compelling, illustrating that the model we've developed achieves outstanding performance benchmarks. Compared to the most recent research findings in academic literature, our model shows higher effectiveness and superiority in the music genre classification task.