Research On Music Genre Classification And Generation Technology Based On Deep Learning

Since ancient times,music has not only been one of the arts continuously pursued by human beings,but also the language of all human beings,so the classification and identification of music is of great research value.With the rapid development of computer science and technology,people began to use computer technology and signal processing technology to conduct research on music,thus forming the field of music information retrieval(MIR).For problems in this field,such as music style classification,algorithmic composition,etc.,many scholars have obtained rich research results from different perspectives.This paper will study music stream recognition and algorithmic composition based on in-depth research techniques.The upper layer of the single neural model is artificially combined with the network manufacturing capability verification learning model,and the neural network test feature learning algorithm after the second artificial neural network experiment is added to form different combined model training,and finally the voting model is the best in this paper model.For algorithmic composition system,based on the timing and complexity of music signal sequence,this paper proposes MSVAE(Music Spectral Variational Auto Encoder)to generate music.It consists of two parts,one is the encoder E(Encoder),which uses a convolutional neural network to generate music similar to the original audio;the other is the discriminator D(Decoder),this paper uses KL(Kullback Leibler)divergence to judge the real Whether the data distributions of music and generated music are similar,and reconstruction error as an overall objective.At the same time,this paper establishes a music generation evaluation system,and compares the generated music samples with the original music from the perspective of audio feature indicators to verify the feasibility and effectiveness of MSVAE to generate music.