The Research On Method Of Feature Extraction For Waveform Music Files

Music is one of the most important forms of the multi-media, which has affectedpeople’s lives widely. As a result, the research and application of the computer musichas attracted lots of attentions. As the basic of the music classification, music retrievaland automatic choreography, the feature extraction of music is highly valued by theresearchers. On the basis of that, a detailed analysis and research on the technology ofwaveform’s feature extraction has been presented in this paper.On the basis of predecessors’ research, this paper proposed a methed of featureextraction for waveform music files according to the characteristics of the signal ofwaveform music files. The research on methed of feature extraction for waveformmusic files include four parts:First of all, this paper presents a method for note extraction, which improves theextraction accuracy rate. Through the analysis of the waveform music files, we getthe information of sampling data and others. Then through the process of signalpreprocessing, gaussian low-pass filter,we get the envelope. Through the FFT to theenvelope, we get the peak of the signal and pick up all the notes successfully. Andthen through the analysis of the notes, we get the features of pitch, intensity andduration.Secondly, we give the method of bars division. According to bars’ characteristicsin music performance on signal, we extract the bars’ line using the methed of addingwindows statistical. Then through the analysis of bars, we get the average andstability of pitch, intensity and duration of bars.Thirdly, we give the method of periods division. We take advantage of thesimilarity between bars to divide the periods, then get the feature vector of peroidswhich can present the changes of music emotion, such as melody, dynamic, rhythm,tempo etc.Finally, we improved the RAN neural network according to the music characters,and then recognited the emotion of music. The feature vector of periods will be inputas the input of RAN nuetral network to recognite the emotion of peroids. Then we get the emotion vector of music according to the Hevner emotion model. Then determinthe emotion of music.Finally, through the experiment, this paper determine that the method ofwaveform music feature extraction has good recognition effect.