Main melody extraction from polyphonic music based on frequency amplitude and multi-octave relation. (March 2021)

• Record Type: Journal Article
• Authors: Li, Chen; Liang, Yajun; Li, Hongmei; Tian, Lihua
• Is Part Of: Computers & electrical engineering. Volume 90(2021)
• Publication Date: 2021-03
• Subjects: Main melody extraction -- Music information retrieval -- Polyphonic music -- Pitch contours; Computer engineering -- Periodicals; Electrical engineering -- Periodicals; Electrical engineering -- Data processing -- Periodicals; Ordinateurs -- Conception et construction -- Périodiques; Électrotechnique -- Périodiques; Électrotechnique -- Informatique -- Périodiques; Computer engineering; Electrical engineering; Electrical engineering -- Data processing; Periodicals; Electronic journals; 621.302854
• DOI: 10.1016/j.compeleceng.2021.106985
• Languages: English
• ISSNs: 0045-7906

Abstract

Abstract: Most melody extraction methods focus only on the temporal continuity of the frequency, which would result in the poor effect of melody extraction for many algorithms. This paper proposes a new main melody extraction framework based on a new salience function. Firstly In the multi-pitch extraction stage, a new salience function combining the frequency and frequency amplitude continuity between adjacent frames is proposed to select fundamental frequency points. By this way, we can not only retain the latent fundamental frequencies at a maximum, but also reduce the number of pseudo fundamental frequency points caused by noise and other factors. Secondly in the melody selection stage, the contours formed by the obtained fundamental frequencies are grouped based on the multi-octave relation and the contour length. For the contours with a multi-octave relation, to reduce the errors caused by the octave problem, the only candidate contour is selected according to the salience and amplitude of the pitch points in each frame of each group. These contours are then merged with the contours without multi-octave relations. Subsequently, the main melody line of each frame is again determined by the salience of pitch points, and is corrected according to the pitch interval progression relation. Some evaluation experiments are carried out to analyze the proposed method, and the experimental results show that the performance of the proposed method is better. Graphical abstract: Highlights: A salience function combined frequency and its amplitude continuity is proposed. Obtained contours are grouped based on multi-octave relation and contour length. One candidate contour is selected according to salience and amplitude in a frame. The main melody line of each frame is determined by the salience of pitch points.