Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Prosody production after stroke: acoustic features analysis on a longitudinal speech corpus
Poster Session A - Sandbox Series, Thursday, October 24, 10:00 - 11:30 am, Great Hall 3 and 4
This poster is part of the Sandbox Series.
Giada Antonicelli1,2, Karen Arellano1,2, Monike Egia2, Simona Mancini1,3; 1Basque Center on Cognition Brain and Language, 2University of the Basque Country, 3Ikerbasque, Basque Foundation for Science
Prosody is the complex of acoustic phenomena in speech. Neurotypical speakers can effectively use it to convey and interpret both linguistic and emotional meanings. Pathological populations, however, often experience a disruption in prosody processing. Studies analyzing speech production observed reduced pitch and envelope variability following brain damage but it remains unclear whether lesions to either hemisphere can differentially affect expressive prosody. In this ongoing study, we examine spontaneous speech recorded from stroke survivors and healthy individuals. We ask (1) which acoustic features best discriminate among groups; (2) what is the effect of lesion and lesion lateralization on prosody; and (3) whether stroke survivors show recovery over time. We analyzed recordings from 3 groups of participants, one with right-hemisphere damage (RHD, N=38; 16 F; Mage = 66.7; SD = 10.6; range = 42-85), one with left-hemisphere damage (LHD, N=48; 18 F; Mage = 68; SD = 10.9; range: 44-89), and a control group with no brain damage, matched in age, sex and education with the two clinical groups (N=15; 4 F; Mage = 59.1; SD = 8.1; range: 39-71). Twenty-four stroke survivors (13 RHD and 11 LHD) were followed up longitudinally at four different time points (TPs): 1) in the acute phase (~1 week post-stroke); 2) in the early subacute phase (3-4 weeks post-stroke); 3) in the later subacute phase (3 months post-stroke), and 4) in the chronic phase (3-6 months after stroke). To allow for a direct comparison, the control group was also tested at four time points following the same timeline as the clinical groups. After preprocessing the audios, we extracted mean pitch (Hz), pitch slope (Hz/s), pitch standard deviation, mean envelope, and envelope standard deviation. We used Kruskal-Wallis and Wilkoxon tests for cross-sectional analyses at TP1 and aligned rank transform models for longitudinal analyses. At TP1, we found an effect of group. Pitch-related measures were higher in the control group relative to both LHD and RHD but no differences between the two clinical groups were found. No statistically significant effect of group was found in envelope-related parameters. Similarly, longitudinal analyses revealed an effect of group for pitch- but not for envelope-related parameters. Finally, no effect of timepoint emerged, although LHD and RHD displayed a positive trend in all features except for mean envelope. These results suggest that clinical and control groups can be effectively teased apart based on speech acoustics such as pitch-related parameters. Acoustic features extraction has therefore the potential to detect prosodic deficits in non-neurotypical populations. However, a larger data set that includes speakers with more severe language and prosodic impairment is necessary to increase its diagnostic power, and to detect potential differences between RHD and LHD speakers, both in the acute and in subsequent stages of post-stroke recovery. Data collection is currently ongoing. We conclude that acoustic analysis of speech can be a quick, sensitive, and objective way to screen prosody after stroke.
Topic Areas: Prosody, Disorders: Acquired