Comparative Study Of Singing Voice Detection Methods

Abstract

This paper studies the detection of singing segments using various features, such as MFCC (Mel Frequency Cepstral Coefficients) and LPCC (Linear Predictive Cepstral Coefficients), with the HMM (Hidden Markov Model) models. The audio clips under test in this paper include isolated segments from different sound tracks and all segments entirely from a sound track. In the literature, these two cases are usually individually investigated. However, we have a unified treatment to both types of segments using the same features and classifiers. In the experiments, we design five experiments to fully examine the performance limitation of the approaches studied in this paper.