Discriminant analysis of the utterance state while singing

Abstract

A considerable amount of contemporary Japanese pop music includes high tones that cannot be sung using only one vocal register. Indeed, attempting to do so often results in throat strain, a situation that often occurs in karaoke. Previous research has focused on evaluating trained, classical, and operatic styles of singing. There is insufficient research on automatic evaluation of the quality of high tones. We have developed a system with 25 dimensions that automatically determines a user's voice quality by identifying his/her vocal register and then recommends a more suitable one. The analysis primarily used fundamental frequency, formant, harmonics, and residual signal. An electromyogram was used to analyze a “tightening voice” in modal voice, which sounds strained and/or represents suffering. The identification rate of the state of utterance while singing was 91.7% per unit of note.