Application of Long Short-Term Memory Intelligent Algorithm in Automatic Classification System

This research focuses on music genre classification (MGC) and music genre recognition within the field of music information retrieval. Specifically, an MGC system is devised leveraging long short-term memory (LSTM) and recurrent neural network architectures. The LSTM model facilitates learning continuous frame feature representations and assimilating statistical information from each segment. The proposed model is evaluated using the GTZAN dataset. Results demonstrate that the LSTM-based MGC system achieves a classification accuracy of 49.8% across 10 music genres, surpassing the baseline convolutional neural network model by 3.13%. Thus, the efficacy of the deep LSTM algorithm in MGC is substantiated. The selection of LSTM for MGC enhances the learning of long-term dependencies and is particularly suited for music signal processing. Moreover, adjustments to time-based gradient back-propagation mitigate issues related to gradient vanishing and explosion.