Research On Music Genre Classification Of MIDI Base On Deep Learning

With the advent of the digital music era,digital music resources are boomed.Music classification is the basis for managing massive music resources.Music genre is an important label for describing music.Music labels play a good role in marking and dividing digital music resources.To classify a huge music database by manually labeling,it will take a lot of cost and time,which can no longer meet the need of the times.Therefore,music automation classification has gradually become a research hotspot.MIDI(Musical Instrument Digital Interface)is an important digital music format widely used in music creation and education.The MIDI music genre classification task consists of two important steps: feature extraction and music classification.Based on the MIDI music genre classification task,this paper improves the feature extraction and classifier design,and proposes a music classification method based on deep learning.The main work and innovations of this paper are as follows:(1)In the traditional classification method,it is difficult to describe the timing characteristics of music for extracting global features of MIDI music.Therefore,in the process of feature extraction,this paper divides the MIDI music into a number of sections with similar local playing styles,and uses the music section as the analysis unit to extract the musical features of the music segments to form a musical feature sequence to better describe the music.The specific process includes extracting the note feature matrix,extracting the main melody and the music segmentation based on the note feature matrix,and researching and extracting the effective features based on the main melody of the segment to form the segment feature sequence.(2)Aiming at the fact that the classifier in the traditional classification method is difficult to learn the timing and semantic information of music due to the limitation of shallow structure,this paper proposes a MIDI music classification method based on deep learning.According to the input sequence of segment features,we studied recurrent neural network and attention mechanism,and use the Bi-GRU(Bi-directional Gated Recurrent Unit)and attention mechanism to design a MIDI music classification network model for the first time.Bi-GRU is good at processing sequence data.It can learn music context semantics and deep features from the segment feature sequence.The attention mechanism automatically assigns different attention weights to the characteristics learned by Bi-GRU from different musical segments,and learns more significant musical features,thereby improving the accuracy of classification.(3)1920 MIDI files with genre tags were collected from the Internet to build datasets for music classification experiments.In the experiment,the music classification method proposed in this paper is used to obtain 90.1% classification accuracy,which is better than the traditional BP neural network-based classification method.Combined with the accuracy of the classification experiment of the equal-length segmentation,the validity of the music segmentation method for music classification is verified.