Acoustic analysis of speech produced with degradation of acoustic and somatosensory feedback

Abstract

The sensory feedback used to control the accuracy of real-time speech production is not always reliable. Noise, occlusion, or experimental manipulations may interfere. State feedback control models of speech motor control predict that speakers will decrease their reliance on feedback when it becomes noisy or unreliable, down-weighting the effects of error. The details of this down-weighting, however, still need to be demonstrated empirically for speech. The present study manipulates speakers’ real-time feedback so that it is simultaneously unreliable in two sensory domains: audition and oral somatosensation. Acoustic feedback was degraded via eight-channel real-time cochlear implant simulation, while an over-the-counter oral anesthetic (benzocaine) was applied to speakers’ tongues and lips to degrade somatosensory feedback. Speakers (N = 18) produced 139 isolated English words under both baseline and feedback-degraded conditions. F1 and F2 measurements were taken from stressed tokens of seven vowels [i,І,e,ae,α,Λ,u]. Significant differences in speakers’ responses to feedback degradation were observed both across vowels and across individuals. Reduction of F2 contrast was the most consistent response observed and similar isolated reduction of vocalic contrast was common across talkers. Hyperarticulation of contrast was also observed in some talkers, however, suggesting that multiple behavioral possibilities exist for situations of bisensory degradation.The sensory feedback used to control the accuracy of real-time speech production is not always reliable. Noise, occlusion, or experimental manipulations may interfere. State feedback control models of speech motor control predict that speakers will decrease their reliance on feedback when it becomes noisy or unreliable, down-weighting the effects of error. The details of this down-weighting, however, still need to be demonstrated empirically for speech. The present study manipulates speakers’ real-time feedback so that it is simultaneously unreliable in two sensory domains: audition and oral somatosensation. Acoustic feedback was degraded via eight-channel real-time cochlear implant simulation, while an over-the-counter oral anesthetic (benzocaine) was applied to speakers’ tongues and lips to degrade somatosensory feedback. Speakers (N = 18) produced 139 isolated English words under both baseline and feedback-degraded conditions. F1 and F2 measurements were taken from stressed tokens of seven vowels [i,І,...