Quantitative evaluation of singing performance using acoustic analysis

Abstract

peoples’ attitudes towards consumption of music have been changed via different ways of enjoying music. Therefore, singer's voice can be treated as the most important element in music. Expressive singing voices vary in terms of tone quality, range, and even skills among singers due to their diverse features. In the area of Music Information Retrieval (MIR), extraction methodologies of fundamental features, such as energy, pitch, formant and modelings of expressive features have been studied, in which singing voice has came into the spotlight. Ultimately, many vocal related studies spare efforts to interpret and evaluate the singer’s voice in diverse perspectives and to find whether this process has correlation and similarity with the qualitative evaluation procedure done by human ears. This study presents the extraction of diverse features related to singing voice and the process of automatic evaluation on singing skills. Although the singers are given to sing with the controlled reference, score or MIDI file, each singer’s characteristics are expressed via performance in terms of habits and techniques. The analyzed features for singing voice are Pitch Range, Pitch Difference between MIDI and Audio, duration and parameters of vibrato sections and more. Such evaluation procedures and itself may be applied in fields of vocal summary, automatic evaluation, musicological analysis in musical education and entertainment. Later, people listened to the singing audio and given to evaluate the vocals in various aspects and investigated the similarity and relationship between the quantitative and qualitative procedures.

People, in these days, not only just listen to the songs, but also sing along the melody