Melody Extraction from Music: A Comprehensive Study

Abstract

Melody extraction plays a significant role in the field of Music Information Retrieval (MIR). Over the past decade it has emerged as one of the active research problem in MIR domain. Nowadays, the music providers have to facilitate searching of music based on their contents or recommend music based on users interest having identical contents. Melody extraction is mandatory to fulfill these user-interest driven searching and recommendation. The primary objective of melody extraction is to achieve a frequency series that corresponds to the pitch of the dominant melody in an audio sample. Numerous approaches have been introduced for melody extraction, mainly from polyphonic music. Melody extraction approaches can be classified into three categories on the basis of the key concepts used for the algorithmic design, namely salience based, source separation based and data-driven approaches. Salience-based approaches have used the salience function of pitch candidate, which is nothing but a pitch salience based on time-frequency representation. Source separation-based approaches distinguish the melody source from the polyphonic music. Data-driven approaches are new and recent development for melody extraction, which basically classify the peak of the pitches of polyphonic music. In this chapter, these approaches are discussed broadly along with the inherent challenges that remains to be solved. In addition, the commonly used datasets and the performance measures that are used to evaluate the performance of melody extraction approaches are also discussed. The MIR applications and the music analysis, where melody extraction plays a vital role, are briefly explained.