Research Of Music Genre Classification Based On Acoustic And Musical Features

Music genre automatic classification is to classify digital music samples by using themethods of signal processing and pattern recognition which is carried out in accordance withthe genre styles by computer, automatic music retrieval and music genre classification hasbecome a hot research direction in recent years. The music genre classification is a difficultpattern recognition problem, meanwhile, is of a great deal of research and application value.In this paper, several feature extraction methods are studied based on the acoustic and musicfeatures to improve the music genre classification accuracy, the specific content is as follows:1、First, the music beats features extraction method is studied, beats semantic featuresand MFCC acoustic features are combined to improve music genre classification rate.Because the strength, speed, duration of music rhythm can reflect the music importantsemantic features of different genres, and beats are the low frequency part produced bypercussion instruments, so six layers of wavelet decomposition is used to extract thelow-frequency beat of music signal. In view of the music tempo difference is not obvious insome, combination of MFCC acoustic features and the beats features are utilized, and basedon the analysis of the mechanism of music genre8-order MFCC is used instead of thecommonly12-order MFCC. The experiment results evaluated on eight genre classes showthat, the overall classification accuracy rate can reach68.4%, while the increase in featuredimensions have little influence in time-consuming.2、Modulation spectrogram features of the music genre classification method based onspectrogram separation are studied. Extracting features from the music samples directly willbe affected by the interaction between the two components. As the rhythm and harmonypresenting different distributions in the time-frequency plane, the percussive and the harmoniccomponents can be separated by applying filtering on the spectrogram. Modulating thepercussive and harmonic spectrograms respectively and then get the music rhythm and rhymemodulation spectrogram features, which describe the long-term mid-level characteristics.Experiment results show that the music rhythm and rhyme characteristics seem to be moreclear after the percussive and harmonic spectrogram separation. And the classificationaccuracy is73.5%for8music genres.3、Music genre classification method based on multi-scale Gabor image texture feature isstudied. To extract acoustic features from multiple mixed voice signal is difficult because theeffects of other elements may make the classification performance worse. Music semanticelements in the spectrogram show a clear visual texture information, spectrogram texturedensity, direction indirectly reflects genre characteristics. So new feature is proposed byextracting spectrogram image multi-scale Gabor texture feature in different angles from theperspective of image processing, the experimental results show that multi-scale Gabor imagetexture feature’s classification performance can match the acoustic features, the overallclassification accuracy is73.1%, the highest can amount to83.3%.