Automatic recognition of temporal speech features in type 2 diabetes mellitus with mild cognitive impairment

Abstract

There is strong evidence of the increased risk for mild cognitive impairment (MCI) among type 2 diabetes mellitus (T2DM) patients. Since slight changes in speech can indicate cognitive impairments, the main aim of our study was to examine the characteristics of T2DM patients’ speech. Participants with T2DM (all above the age of 50) were divided into two groups: patients with MCI (n =26) or with normal cognition (n = 23). Spontaneous speech samples (containing the subjects’ description of their previous day) were collected and used for extraction of temporal speech parameters via an automatic speech recognition (ASR) based tool. Results indicated longer and more frequent filled pauses in MCI-patients’ speech: frequency of filled pauses (p<0.004), ratio of filled pauses and phonemes (p<0.038) and the length of filled pauses compared to total utterance length (p<0.022) showed a statistically significant difference between the groups. Automatic analysis of spontaneous speech could be a promising, non-invasive way of early detection of cognitive decline in T2DM in the future.