Research On Music Style Classification Of MIDI Base On Musical Segment Features

MIDI,as a storage format for music,have received extensive attention because of its advantages of small size and easy editing.Music style is an important feature of music,and it is also a label for music sites to distinguish music files.Most of the previous music style classification methods use manual labeling methods for classification.Nowadays,more and more MIDI files are on the Internet.Labeling a large number of documents one by one will consume a lot of manpower and time.Therefore,solving the MIDI music style classification problem through artificial intelligence has become a new topic of machine learning.MIDI music classification research can be roughly divided into two links.One is to extract the music features that can represent the music from the file,and the other to use these features to train a suitable classifier.The current research basically uses the statistical characteristics of the MIDI main melody as the music features and uses the BP neural network as a classifier.However,in the main melody,the order of the notes has a great influence on the characteristics of the melody,and the statistical characteristics cannot fully reflect this relationship of the main melody.Second,these features do not include music accompaniment information.In addition,a piece of music often has many repeating melody segments.If the entire piece of music is taken as a sample,it will increase the redundancy of the sample.In view of the deficiencies of previous methods,this paper studies the MIDI music classification based on the music segment features.The main research work and innovations of this paper are:(1)This paper proposes a method for extracting the musical segment features from a MIDI file.This method divides each piece of music into a number of sections and extracts the musical segment features of each section.The specific process of this method includes MIDI file note extraction,main melody extraction,music segmentation and feature extraction.Musical segment features are extracted from musical segment of music,contain the main melody and accompaniment information of the piece,and can reflect the order relationship of the notes.(2)In-depth study of the recurrent neural network,and studied an improved model of it,the GRU neural network,and based on its advantages in processing sequence data and overcoming the long-term dependence problems in the recurrent neural network and the sequence properties of the feature,we decided to use GRU as the classification network of this paper,and a GRU-based MIDI-style classifier was designed.(3)Establish a complete MIDI music style classification model,verify the accuracy of the model,and compare the experimental results with previous methods.The results show that the classification model proposed in this paper has higher accuracy than the existing results.