A Research On Music Emotion Analysis Based On Feature Vector

With the quick informatization in today’s society, all kinds of multimedia information developing rapidly. As an art form, Music has become a necessary part in human life. It has always been that music is a channel for people to express their feelings. We can sing for happiness and sadness. Now the music on the paper already can not satisfy the music storage, retrieve and communication between the musicians. With the coming of information age, computer music research has become a new topic. Let the computer to finish what we can accomplish has been our direction. At present, through the computer we can do music broadcast, production and storage, etc. The analysis of music emotion through the computer also gradually rise. Make the computer can identify music emotion automatically by "listening to" music. In this paper, the music emotion automatic analysis done in-depth research.Music emotion analysis model in this paper consists of three parts: the music feature vector model, music emotion model and classify cognitive model.Music feature vector model is an eight-dimensional vector by extracting some of the features from the music composition. In the part of the music feature vector model, this paper introduces definition of the concept of musical energy after the concept of melody area, and made our own way, that is use musical energy of the music divide the music to some sections, and for each segment, using digital music feature extraction technology to extract music section’s speed, the direction of melody, dynamics, tempo, tempo changes, major third, minor third features, then use music emotion model and classify cognitive model to analyze each music segment.Music emotion model is a description of musical emotion, this paper introduces several musical emotion model researchers used, including Hevner emotion ring, Thayer emotion model and emotional semantic model, etc., and the advantages and disadvantages of these models were compared. We combined Hevner emotion ring and emotion semantic model, and then the emotion vector model obtained which is composed of eight categories describe in Hevner emotion ring, and the model as affective model used in the experiments herein.Categories cognitive model is mapped characterized model to the music emotion model by the algorithm, namely the classification of cognitive process is a pattern recognition process. in the classification cognitive model section, after a brief introduction on several pattern recognition methods and their advantages and disadvantages were compared, BP neural network is chose as a cognitive model of this paper. Demand for music emotion analysis, improvements of BP neural network learning process have been made in this article, so that it can be more in line with the characteristics of musical emotion subjective analysis.Finally, the three parts naturally combined to form a complete musical emotion analysis models. After that, the functionality and performance of the model experimentally verified, and comparing the experimental results with the experimental results of previous studies, the result shows that musical emotion analysis model using the proposed method can build a better digital music emotional analysis, and compared with the existing achievements, it has a higher accuracy rate.