On the effect of artificial distortions on objective performance measures for dialog enhancement

The objective evaluation of dialog enhancement systems using computational methods is desired to complement the subjective evaluation using listening tests. It remains a challenge because for this application neither were performance measures specifically designed, nor were existing measures systematically analyzed. This work investigates eight objective performance measurement tools originally developed for audio and speech coding, speech enhancement, or source separation. To this end, a set of basic distortions is presented and used to simulate degradations that are common in dialog enhancement. The effect of the artificial distortions on the performance measures is quantified by means of a so-called response score that is proposed here.