Presentation Information
[P-8-07]Quantifying Speech Slowing through Prosodic Analysis of Episodic Memory Recall in Alzheimer's Disease
*Miju Kang1, Sungmin Kim1, Suji Jung2, Jiwon Chun2,3, Dajin Kim4 (1.Department of Medical Sciences, The Catholic University of Korea College of Medicine, Seoul, Republic of Korea(Korea), 2.Department of Medical Informatics, The Catholic University of Korea College of Medicine, Seoul, Republic of Korea(Korea), 3.The Catholic Medical Center Institute for Basic Medical Science, The Catholic Medical Center of The Catholic University of Korea, Republic of Korea(Korea), 4.Department of Psychiatry, Seoul St. Mary’s Hospital, The Catholic University of Korea College of Medicine, Seoul, Republic of Korea(Korea))
Keywords:
Alzheimer's disease,Episodic memory recall,Prosodic features,Spontaneous storytelling
Background: While previous dementia detection studies primarily used picture description tasks, these structured tasks differ significantly from spontaneous speech patterns, potentially limiting their applicability to real-world conversational settings. This study analyzed prosodic features from a Korean speech dataset (AI-Hub) comprising multiple tasks, focusing on spontaneous storytelling from healthy controls (HC) and Alzheimer's disease (AD) patients with validated clinical scores.
Methods: From a comprehensive dataset of 1,002 participants, HC (n=30) and AD (n=29) were randomly selected for this exploratory study. Audio files were transcribed using OpenAI-Whisper model to obtain speech-to-text conversion and timestamps. Speech data were collected using a custom-developed application across 11 hospitals, with clinical information validated by primary clinicians and specialist physicians. We analyzed the "storytelling" task, where participants recalled yesterday's events for one minute, as it most closely resembles spontaneous speech. Prosodic features extracted included speech rate, phonation percentage, articulation rate, and syllable duration. Descriptive statistics were computed to assess speech dysfunction patterns between groups.
Results: Cognitive assessments confirmed group differences: MMSE scores (HC: 28.2±2.1 vs AD: 21.2±4.0) and CDR scores (HC: 0.05±0.15 vs AD: 0.8±0.5). AD patients showed significant impairments across all prosodic measures. Speech rate decreased by 27% (HC: 2.75±0.80 vs AD: 2.01±0.98 syllables/sec), articulation rate declined by 21% (HC: 2.95±0.77 vs AD: 2.33±0.95 syllables/sec), and phonation percentage was reduced (HC: 92.6±6.4% vs AD: 85.8±16.7%) with increased variability. Syllable duration was prolonged by 44% in AD (HC: 362.5±97.4ms vs AD: 522.9±275.5ms). Language production was significantly diminished, with syllable count and word count reduced by 30% and 26%, respectively.
Conclusion: Prosodic analysis demonstrates significant speech deficits in AD patients, suggesting that speech features represent promising objective biomarkers for dementia diagnosis and monitoring in naturalistic settings.
Methods: From a comprehensive dataset of 1,002 participants, HC (n=30) and AD (n=29) were randomly selected for this exploratory study. Audio files were transcribed using OpenAI-Whisper model to obtain speech-to-text conversion and timestamps. Speech data were collected using a custom-developed application across 11 hospitals, with clinical information validated by primary clinicians and specialist physicians. We analyzed the "storytelling" task, where participants recalled yesterday's events for one minute, as it most closely resembles spontaneous speech. Prosodic features extracted included speech rate, phonation percentage, articulation rate, and syllable duration. Descriptive statistics were computed to assess speech dysfunction patterns between groups.
Results: Cognitive assessments confirmed group differences: MMSE scores (HC: 28.2±2.1 vs AD: 21.2±4.0) and CDR scores (HC: 0.05±0.15 vs AD: 0.8±0.5). AD patients showed significant impairments across all prosodic measures. Speech rate decreased by 27% (HC: 2.75±0.80 vs AD: 2.01±0.98 syllables/sec), articulation rate declined by 21% (HC: 2.95±0.77 vs AD: 2.33±0.95 syllables/sec), and phonation percentage was reduced (HC: 92.6±6.4% vs AD: 85.8±16.7%) with increased variability. Syllable duration was prolonged by 44% in AD (HC: 362.5±97.4ms vs AD: 522.9±275.5ms). Language production was significantly diminished, with syllable count and word count reduced by 30% and 26%, respectively.
Conclusion: Prosodic analysis demonstrates significant speech deficits in AD patients, suggesting that speech features represent promising objective biomarkers for dementia diagnosis and monitoring in naturalistic settings.