The perception of synthetic speech versus natural speech stimuli in adult listeners

Abstract

The majority of speech perception research investigates interpersonal perception through presentation of two types of stimuli: synthetic speech and/or natural speech. Synthetically created speech stimuli allows for precisely controlled manipulation of individual acoustic variables, but can sound unnatural calling into questions its ecological validity. Alternatively, natural speech preserves the full complement of redundant acoustic cues yet limits control of specific acoustic parameters. In this study, we investigate speech discrimination comparing perception of specified acoustic parameters (formant transition duration, F3 onset frequency, F3-F2 distance) across three types of speech stimuli: synthetic speech stimuli, natural speech of a standard adult speaker, and each participant’s own voice. We use the Wide Range Acoustic Accuracy Scale (WRAAS), a computer-based perceptual assessment that uses a parameter estimation by sequential testing (PEST) adaptive-tracking algorithm to rapidly identify a listener’s ability to discriminate sounds. Most trials are presented near the Just-Noticeable-Difference (JND) of a stimulus continuum; thus, few trials are “wasted” with obviously same or different sounds. This study of typical adult listeners (n = 20, enrollment is ongoing) will provide baseline data for future work investigating perception of these same stimuli in children.