Music Genre Recognition Research Based On Improved Deep Convolutional Neural Network

The genre expression of audio features and the design of feature extraction in the field of music genre classification seriously affect the accuracy and generalization of the classification method.Therefore,a DCNN-AFC model is proposed of dual attention fusion extraction features,and Enhance the genre expression of the features in the audio power spectrogram.Firstly,to consider the diversity of music genre features in the audio power spectrogram,the Mel spectrum method is used in the feature extraction stage to effectively filter the audio signal by simulating the human ear auditory system,and the dimensionality of the signal after Mel filtering is restored.To ensure the effective retention of the genre feature attributes of the audio signal,and deepen the feature difference extracted between different genres.To reduce the input size and expand the model training scale in the model input stage,the audio power spectrogram obtained by feature extraction is cut into 128×128×1 and input the model,which further improves the calculation speed and training effect of the model.Secondly,The model’s deficiencies in the design of genre feature extraction are designed to design a residual module for deep learning that replaces the original ordinary convolution operation with a residual structure and enhances the depth of the network.While enhancing the highly abstract extraction of genre features,it avoids model degradation such as gradient disappearance.The problem is to ensure the effectiveness of the model;Finally,the channel attention mechanism and spatial attention mechanism are merged,and the genre features of the audio power spectrogram are coordinated and calibrated in the channel domain and the spatial domain,and the features extracted by the dual attention fusion Information input residual module,through the residual characteristics in the module to describe the features extracted by dual attention fusion in detail,improve the differential expression of features between different genres,thereby enhancing the directionality of genre feature extraction,and then improving the classification of music genres effect.Experiments show that the DCNN-AFC model is superior to other commonly used deep learning models in enhancing the feature extraction effect of music genres and the efficiency of genre classification.It also improves the accuracy of music genre classification by 6.27% to 11.35% compared with other deep learning models.Verify the effectiveness and advancement of the DCNN-AFC model.There are 37 figures,12 tables and 54 references in this paper.