Singing voice quality assessment in professional singers using microphones and contact sensors

This study focuses on objective evaluations of singing voice quality obtained from the analysis of different singing tasks. Three devices were used to simultaneously acquire the performed tasks, thus allowing a comparison among the estimated parameters to be obtained. Fourteen professional singers took part in the experiment: they sang two Italian songs at comfortable tonality and performed an arpeggio using the vowel /a/, while standing in front of a sound level meter (SLM) and wearing two types of contact sensors, namely an Electret Condenser Microphone (ECM) and a Piezoelectric Contact Microphone (PM). They also read aloud an Italian phonetically balanced text. The singing voice quality was assessed by means of parameters related to pitch inaccuracy and singer’s formant. In addition, Cepstral Peak Prominence Smoothed (CPPS) distributions were investigated in songs for the first time. The pitch inaccuracy estimation was comparable for the three devices: the overall mean of the pitch deviation between each contact microphone and SLM was equal to 1.9 Hz (standard error: 0.4 Hz), thus making ECM and PM as good as SLM for this estimation. Significant differences were found for the singer’s formant evaluations, which were performed observing the Long-Term Average Spectra (LTAS) and calculating the Singing Power Ratio (SPR): since the signal acquired at the output of the contact sensors is affected by the physiological filtering (vocal folds – throat – skin), but not by the filtering effect of the vocal tract that affects microphones in air, the two contact sensors showed a higher spectral slope. However, a good correlation between SPR obtained from ECM and SLM proved contact microphones are able to highlight the presence of the singer’s formant. Furthermore, evidence has been found that CPPS distributions shape from the three types of microphone indicate the degree of singing voice quality.