An Automatic Singing Evaluation System

In this paper, a singing evaluation approach based on pitch and rhythm accuracy is proposed. A comparative study was first carried out between speech and singing voice, which demonstrated the characteristic of consonant and vowel in singing voice. A phoneme boundary detector was adopted to perform note segmentation in a high time resolution by using characteristics of consonant and vowel. As for singing skill evaluation, relative pitch interval between reference score and actual sung pitch was adopted to estimate pitch deviation while difference between absolute position and expected position of beats served as rhythm consistency measure. Experiments on a solfege corpus demonstrated a noticeable improvement in the performance of onset detection. A detailed subjective evaluation showed great consistency between the proposed evaluation approach and experts’ judgment.