Voice Analysis Method Based on Deep Hybrid Network

Abstract

In modern society, all kinds of music emerge in endlessly. With the popularity of music, people's requirements for music signal processing and analysis technology also increase. Different audio processing, music retrieval and other technologies have been proposed. Based on the deep hybrid network model, this paper analyzes timbre, one of the basic elements of music. This paper designs and implements a voice analysis system based on pattern recognition technology. First of all, according to the obtained timbre feature parameter vector, combined with pattern recognition technology, different classifiers are used to classify different musical instrument sound effects. Secondly, different classifiers are trained based on data mining platform. Different classification models are generated and the classification accuracy of different classification models is compared. Then, the classification results of different classification models are compared and analyzed, which lays an algorithm foundation for the design of voice analysis system. Finally, based on Java and MYSQL, combined with BP neural network algorithm, a voice analysis system with self-learning ability is designed and implemented. The system realizes the functions of music spectrum component analysis, timbre feature parameter extraction and instrument timbre classification and recognition.