웨이블릿 허스트지수 기반의 음악 장르 분류에 관한 연구

Abstract

As influence and interest in smart phones and cloud internet service is increasing, the interest in digital content industry such as movie. Music content industry is increasing its business size by every seconds and getting its interest by consumers. It is essential to analyze and control all the data for proper services. Considering music industry, in order to supply the meta-data of services, a direct approach on signal data for classifying genres has been researched. One of the previous methods is using the Mel Frequency Cepstral Coefficients(MFCCs) to estimate features of specific music to sort out its genres or discriminate between speech or music signals. The method has its limitation due to weakness in additive noise. Using the other feature extraction of music signal data would be an alternative method to give better observation to analyze. We used a wavelet transform to decompose the signal data to distinguish between noise and coarse signals. After transform, we applied the Hurst exponents to distinguish three major genres of music data. For the conclusion, collaborative systems between hurst features and other meta-data of signal can provide an effective algorithm for audio signal processing.