Poster D8, Friday, August 17, 4:45 – 6:30 pm, Room 2000AB

Neural mechanisms underlying the impact of speech sound naturalness during transformed auditory feedback

Sadao Hiroya1, Takemi Mochida1;1NTT Communication Science Labs

Articulatory compensations in response to formant perturbation in vowels have shown that auditory feedback plays an important role in speech production. Although most conventional perturbation studies have used linear predictive coding (LPC) for estimating formants, it is known that formant estimation accuracy would be degraded due to a glottal source-vocal tract interaction especially for low vowels and female speech. Misestimating of formants by LPC would result in degradation of sound naturalness of transformed speech. To improve sound quality, we have developed a real-time robust formant tracking system using phase equalization-based autoregressive exogenous (PEAR) model which utilizes the glottal source signals measured by electroglottography (EGG). Our previous results have shown that compensatory responses to formant perturbation using PEAR were significantly larger than LPC, possibly due to the improved speech sound naturalness by PEAR. In this study, to investigate the neural mechanisms underlying the impact of speech sound naturalness, we performed fMRI scans during transformed auditory feedback in which formant frequencies estimated by LPC or PEAR were perturbed. For this purpose, we developed a novel PEAR-based system which does not require EGG, so that it is applicable to fMRI scans. Speech stimuli consisted of eight /Ce/ syllables. Formants were transformed so that vowel /e/ shifted to sound like /a/. fMRI scanning was performed using a sparse sampling protocol. Formant perturbation by LPC or PEAR randomly occurred. Behavioral results showed that compensatory responses to PEAR was larger than those to LPC, consistent with previous studies. fMRI results showed that left and right anterior cingulate cortices (ACC) were more activated for PEAR than for LPC. This probably indicates that natural transformed sound by PEAR resulted in greater activation in the ACC and the larger compensatory responses. On the other hand, left-lateralized cerebellum was more activated for LPC than for PEAR. This implies that less natural transformed sound caused greater cerebellum activation. Relating to the above, recent studies showed that patients with cerebellar degeneration showed a larger compensatory response than control subjects. Thus, the involvement of the cerebellum during auditory feedback will be an issue to be addressed in the future. In conclusion, the impact of speech sound naturalness in transformed auditory feedback emerged in the ACC and the cerebellum.

Topic Area: Speech Motor Control and Sensorimotor Integration