Title: A detection method of similar segments of music based on multi-feature fusion

Authors: Yu Zhou

Addresses: Guangzhou Academy of Fine Arts, Guangzhou 510006, China; Bangkokthonburi University, Bangkok 10170, Thailand

Abstract: In order to improve the problem of the high rejection rate of similar music fragments, this paper proposes a similar music segment detection method based on multi-feature fusion. Firstly, the music audio signal is decomposed into perceptual subspaces to express the music's audio signal features in each subspace, and a scale vector and parameter matrix are used to save atomic and molecular information of the signal. Then, the music beat histogram is extracted by a discrete wavelet transform to obtain its dynamic and static characteristics. On this basis, isolated words in similar music fragments are identified according to the results of multi-feature fusion, and then similar music fragments are detected by the classification of similar feature sets. The experimental results show that the semitones rejection rate of this method is between 1.9% and 3.5%, and the accuracy of the music signal connection is between 92% and 98%.

Keywords: similar music segment; similarity detection; audio signal features; multi feature fusion; beat histogram; isolated words; feature set.

DOI: 10.1504/IJICT.2023.132764

International Journal of Information and Communication Technology, 2023 Vol.23 No.2, pp.137 - 152

Received: 11 May 2021
Accepted: 12 Jul 2021
Published online: 09 Aug 2023 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article