Phoneme Recognition in Popular Music

Automatic lyrics synchronization for karaoke applications is a major challenge in the field of music information retrieval. An important pre-requisite in order to precisely synchronize the music and corresponding text is the detection of single phonemes in the vocal part of polyphonic music. This paper describes a system, which detects the phonemes based on a state-of-the-art audio information retrieval system with harmonics extraction and synthesizing as pre-processing method. The extraction algorithm is based on common speech recognition low-level features, such as MFCC and LPC. In order to distinguish phonemes, three different classification techniques (SVM, GMM and MLP) have been used and their results are depicted in the paper.