Chorus Detection with Combined Use of MFCC and Chroma Features and Image Processing Filters
A computationally efficient method for detecting a chorus section in popular and rock music is presented. The method utilizes a distance matrix representation that is obtained by summing two separate distance matrices calculated using the mel-frequency cepstral coefficient and pitch chroma features. The benefit of computing two separate distance matrices is that different enhancement operations can be applied on each. An enhancement operation is found beneficial only for the chroma distance matrix. This is followed by detection of the off-diagonal segments of small distance from the distance matrix. From the detected segments, an initial chorus section is selected using a scoring mechanism utilizing several heuristics, and subjected to further processing. This further processing involves using image processing filters in a neighborhood of the distance matrix surrounding the initial chorus section. The final position and length of the chorus is selected based on the filtering results. On a database of 206 popular & rock music pieces an average F-measure of 86% is obtained. It takes about ten seconds to process a song with an average duration of three to four minutes on a Windows XP computer with a 2.8 GHz Intel Xeon processor.