Music boundary detection with multiple features

Abstract

Music structural analysis tasks have an important position in the field of Music information retrieval which require an understanding of how humans process music internally, such as music indexing, music summarization, and similarity analysis. Many schemes have been proposed to analyze the structure of recorded music, however they usually use single feature to detect boundaries of songs and the results are not satisfactory. In this paper, we present a method which is based on novelty detection and combines multiple features to the task of music boundaries detection. We extract peaks of novelty function derived from various features as potential boundaries, then eliminate non-boundaries from potential boundaries derived from distinct feature sets. Three types of features, including intensity, timbre, and harmony are employed to represent the characteristics of a music clip. On our testing database composed of 175 entire songs, the best accuracy of boundary detection with tolerance ±3 seconds achieves up to 65.7%.