Structural segmentation of music with fuzzy clustering

A study was conducted to segment a digital music file, such as an MP3 file. The study demonstrated that musical segmentation was performed by using MPEG-7 features and constrained clustering based on Means. A team of researchers developed a method, called RefraiD that defects the chorus sections of music and can detect key changes in choruses, using the 12-dimensional chroma feature vector. The researchers investigated musical segmentation of structural components, using Mel frequency Cepstral Coefficient (MFCC) and compared the sequence approach of structural segmentation with the state approach (HMM). The researchers showed that the state approach is more robust and computationally efficient. A method was also proposed for musical segmentation by detecting boundaries and aggregation.