Music feature extraction based on fractal dimension theory for music recommendation system
Li Bi, Qiang Tao, Xiang Li · 2016
Music feature extraction is widely used in music recommendation system.The recommended music is somewhat similar in the form of melody.This phenomenon exhibits a repeating pattern that between the music set and the recommended one, which reveals that the similar music has the characteristics of fractal dimension.In this paper, we selected time energy, frequency energy, Mel-Frequency Cepstral Coefficient (MFCC) and spectral envelope as the music features.These four features were integrated as the feature vectors that were calculated the fractal dimension by Hilbert transform.Compared with the traditional content-based music retrieval, this method focuses on the certain degree of self-similarity between the whole and the local area.This result shows that the feature extraction approach of fractal dimension provide an effective method for music retrieval.