Voice quality assessment by means of comparative judgments of speech tokens

The presentation concerns a perceptual assessment of disordered speech, which rests on comparative judgments. Pairs of stimuli are presented randomly to listeners who are asked to label the most degraded sample. At the end of a listening session, speech signals are assigned overall scores on the base of individual comparative judgments. To contrast the perceptual assessment of voice disorders via comparative judgments with a conventional method, two groups of listeners have taken part in the experiment. A first group has been comprised of six naive listeners, i.e., listeners without any training in speech therapy or laryngology. A second group of listeners has been comprised of three speech therapists. They rated from 0 to 3 the grade G (of the GRBAS scale) of each sound sample by means of an ordinal scale. Results show that comparative judgments give rise to high intra-judge and inter-judge agreements for sustained vowel [a] as well as for sentences even though the listeners had no prior experience in rating disordered voices.