Identification of novel multidimensional speech-like sounds and the effects of training

Abstract

Listeners classified Klatt synthesized nonspeech stimuli that differed on three independent and discriminable dimensions. The three dimensions of the stimuli were fricative spectrum, frequency transition, and temporal gap. Each of the three dimensions could take on one of three possible values. Twenty-seven stimuli were synthesized using all possible combinations of the dimension values. Multidimensional scaling of paired-comparison similarity judgments confirmed the existence of three perceptual dimensions. Subjects were then trained to classify three exemplar stimuli as ‘‘circle,’’ ‘‘triangle,’’ and ‘‘square,’’ respectively. Following this training, subjects classified all 27 stimuli as ‘‘circle,’’ ‘‘triangle,’’ or ‘‘square.’’ From this it was determined how attention was distributed among the dimensions. Finally, the subjects’ ability to shift their attention from dimension to dimension was examined. [Work supported by NIA and AFOSR.]