Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Are fMRI responses to auditory feedback perturbations during speech linked to speech compensation?
Poster Session B, Wednesday, September 30, 4:30 - 6:30 pm, Wangari Maathai
Matthew H Davis1, Zifei Wang1, Máté Aller1, Dace Apsvalka1, Caitlin Brett1, Abigail R Bradshaw1; 1MRC Cognition and Brain Sciences Unit, University of Cambridge
During speech production, speakers continuously monitor the sensory consequences of their speech. Auditory feedback, in particular, is used to maintain accurate and fluent articulation. However, it is unclear what mechanisms are involved in using auditory feedback. Many accounts propose auditory computations that compare predicted and actual sensory outcomes (ie. prediction error, Guenther & Hickok, 2015). However, whether and how prediction error computations are neurally implemented is still debated (Meekings & Scott, 2021). During speech production, auditory responses are reduced relative to passive listening (speech-induced suppression, SIS). To investigate how SIS is affected by auditory feedback we used fMRI to measure brain responses while participants spoke single words. We assessed two manipulations to auditory feedback: (1) changes to the clarity of speech (using a real-time noise-vocoder) and (2) whether formant frequencies match those intended by speakers (using real-time formant frequency perturbations). Behavioural and neural effects of these forms of mismatch allow us to test the relationship between behavioural compensation for formant perturbation and neural SIS to assess evidence of prediction error computations. We scanned 32 native-English speakers who completed a speech production task with modified auditory feedback in a 3T Siemen Prisma MRI system. Participants produced one of twelve CVC words (head, had, hid, etc) while hearing real-time auditory feedback manipulations using a custom version of Audapter software (Bradshaw, Gualtier, Black & Davis, submitted). Clarity and formant frequency perturbations were randomized in single trials: (1) clear or noise vocoded auditory feedback; (2) unperturbed or perturbed vowel formants with: (i) a 212mel increase in F1 and the same shift down for F2 (F1up condition), or (ii) a 212mel shift down for F1 and up for F2 (F1down condition). These changes produce mismatches of vowel identity for clear and vocoded speech feedback. We collected concurrent fMRI data (TR = 3000 ms, TA = 1034 ms) during speech production using a multiband accelerated sparse imaging protocol so participants spoke and heard auditory feedback during silent interscan intervals. Formant tracking was used to explore how clarity and perturbation influenced F1 and F2 frequencies at the midpoint of the spoken vowel; participants whose speech oppose the changes to formant frequency perturbations provide evidence of behavioural compensation. fMRI analysis compared neural response amplitudes for perturbed and unperturbed vowels to assess release from SIS. Preliminary analyses revealed behavioural compensation to altered auditory feedback. Speakers produced higher F1 frequencies when they heard speech with downward perturbed formants and vice-verse for upward perturbations. Initial results of fMRI group-level analysis showed greater superior temporal gyrus (STG) responses to speech heard with formant perturbations compared to conditions in which expected formants were heard. Planned analyses will compare behavioural and neural responses to clear and vocoded speech feedback to assess whether neural suppression is linked to formant changes for individual participants. These findings provide preliminary evidence that auditory-motor compensation in response to perturbed speech is associated with fMRI response increases. Planned analyses of multivoxel patterns for perturbed and unperturbed vowels in the STG will be used to test for prediction error representations.
Topic Areas: Speech Motor Control, Speech Perception