A psychoacoustic auditory model to evaluate the performance of a voice activity detector

A new psychoacoustic auditory model to evaluate the subjective performance of a voice activity detector (VAD) is presented in this letter. The mathematical model proposed makes it possible to pass from the power spectral density of the speech signal processed by a VAD to an analysis of the subjective loudness density and thus to subjective measures expressed in terms of comparison mean opinion scores (CMOS). In case studies, the correlation between the measured and predicted CMOS values always remained above 0.9, using traditional analytical methods such as regression curves.