摘要：In recent years, machine learning and deeplearning technologies are maturing. The Convolutional NeuralNetworks (CNNs) are applied to all kinds of fields and variousCNN-based fusion and combination methods are also appearedone after another. Due to the streaming media rapid growth,therefore the music genre classification is significant in themultimedia world. In order to raise the user’s efficiency whensearching for different styles of music, we applied CNNcombined with Recurrent Neural Network (RNN) architectureto implement a music genre classification model. In thepre-training step, the Mel-Frequency Cepstrum (MFC) is usedas feature vector of sound samples. We use Librosa to convertoriginal audio files into their MFC to achieve a sensory patternclose to that of humans hear. In this study, a model is trained byMel-Frequency Cepstral Coefficients (MFCC) and CRNNmethod with the accuracy achieve to 43%. This model willcontinue to be improved in the future to identify the music styleby extracting more sound features.
关键词：Convolutional Neural Networks; Recurrent Neural Network; music genre classification; Mel-Frequency
