Automated bimodal pause analysis for acoustic markers of cognitive decline and Alzheimer's disease in connected speech

Abstract

INTRODUCTION: This study introduces a language-independent, acoustic-based method to identify the bimodal pauses in connected speech related to Alzheimer's disease (AD) through a log-normal distribution, aiming to explore pausing behavior as a digital marker of cognitive decline. METHODS: We fitted a bimodal log-normal distribution to 4473 pauses automatically extracted through acoustic analysis. We compared linear and logarithmic pause indices between cognitive groups and explored their neurocognitive correlates. RESULTS: We empirically revealed a dual-mode pause distribution, customizing a threshold of ≈ 180 ms to differentiate short and long pauses. This bimodal distribution effectively distinguished cognitive groups, driven by variations in the central tendency of long pauses. Both pause types were elevated in individuals with mild cognitive impairment and correlated with tau and amyloid levels. DISCUSSION: Bimodal pause distribution shows promise as a sensitive speech-based indicator of cognitive decline, linking closely to AD biomarkers. We introduce a refined, unbiased, language-independent framework for broader application across diverse populations. Highlights: Pausing in connected speech was investigated as a digital marker of cognitive decline. Bimodal log-normal pause distribution distinguishes between cognitive groups. Short (80–180 ms) and long (> 180 ms) pauses correlate with tau and amyloid.