An Analysis of Acoustic Features for Attention Score in Thai MoCA Assessment

Abstract

Screening tests like the Montreal Cognitive Assessment (MoCA) can help diagnose mild cognitive impairment (MCI). MoCA comprises subtests that span various cognitive domains. Numerous researchers attempt to detect MCI by employing speech-related features such as acoustic, linguistic, and prosodic features. However, the features can distinguish patients with MCI from healthy people but do not describe each patient's specific cognitive domain impairment. This study focuses on Digit Backward Span (DBS) and Digit Forward Span (DFS), subtests related to the cognitive attention domain in MoCA. We develop a model and identify the most relevant speech features for the domain from a recorded voice from these subtests in the Thai MoCA. We rank features by their importance and found that using a subset of important features has higher predictive power than using the entire feature set in impairment in the attention domain. The most important features in both tests are the median duration of voice and the duration of voice.