The Research On Method Of Music Emotion Recognition

Music emotion recognition (MER) is a process of using computer to analyze and process music feature. It studies the mapping relations of music feature space and emotional space, and it realizes the cognition of emotion that music expresses. In recent years, with the extensive application of human-computer emotion interactions technology, music emotion recognition becomes a research hotspot.Existing method of music emotion recognition has the following two shortcomings: First, the emotion that music express have changed along the music's playback, and the analysis based on the entire piece of music is difficult to accurately express the ups and downs of music emotion. Second, analyzing the music emotion with the use of pitch, duration, and intensity of note is difficult to reflect the soul and content of music. For these reasons, the paper puts forward a method of music emotion recognition based on the period. Main tasks are as follows:1. According to the organization of music, the paper puts forward a music model. Music is composed of several periods that have independent emotion semantic, and these periods are composed of several sequences of bars, and the bars consist of several notes. The note has three characteristics of pitch, intensity and duration. In order to reflect the ups and downs of music emotion, the paper recognize the emotion of music base on the period. During the study of basic music theory, we extract the bar's feature according to the notes ,and further combine the bars into separate period on the base of the features of the bars, ultimately get the whole piece of the emotion of the music.2. The paper puts forward a method that the music is divided into separate periods. It extracts the pitch, duration and intensity feature of the notes from a MIDI file through analyzing the format of a MIDI file, and it extracts the master track according to the entropy of information of the pitch and intensity. It extracts the feature of the bars according to the notes feature. The music is divided into several independent periods according to the similarities of adjacent music bars, which was judged by the cosine formula with multiple angles.3. In order to understand the essence of music emotion, the paper holds the regular movement of the notes in the music. It extracts the feature of interval, melody, dynamic, meter, rhythm, and tempo and so on, and it recognizes the emotional of the periods by BP neural network. The paper uses the vector to signify the music emotion according to the Thayer emotion model. Ultimately, given the emotion of the entire music though counting up the proportion of the emotional type appeared in the entire music and the proportion of the sum of the period length occupied in the entire music.Experimental results show the method of music emotion recognition based on the period conform to the organization structure of music. It meets the emotional ups and downs, and it has achieved a better effect.