Research On Music Sytle Based On Music Signal Processing

With the development of Internet, digital music has become an important type to record and disseminate music. It is a popular research topic to analyse and process music signal on the computer along with the development of computer science and digital signal processing. A significant research direction is the music classification and recognition based on machine learning. There are many achievements from various perspectives based on different machine learning algorithms.In this dissertation, the performance style of music will be analysed based on signal processing and machine learning. Violin solo music is chosen for the research. Violin is a very popular musical instrument around the world, which is loved by many people because of its rich performance skills and expressiveness. The characteristics of four basic method of violin bowing will be explored, and machine learning algorithms will be applied so that the computer can analyze and recognize these bowing methods. There are two main issues which need to be studied. One is how to get the feature parameters of each method, the other is how to recognize these methods.For the feature parameters, some commonly used feature parameters are firstly anlysed. Then some detailed analysis and comparison of each type are made. 11 specific feature parameters are selected to distinguish the styles according to the characteristics of each type. For the recognition, two kinds of mature machine learning algorithms, the decision tree and the support vector machine, are chosen for training. To get better classification results, large amount of comparison and adjustment of parameters of the algorithm are made. The feature parameters are optimized according to the result of training. Finally, an application system is designed to apply the research results. The simulation results show that with one of the feature parameters removed,the system is able to achieve a cross validation accuracy of 84.5313%under some appropriate parameters of the training algorithm, which indicates that the feature parameters and the parameters in the algorithm designed in this dissertation have access to solve the problem.