Effects of cognitive exercises on speaking characteristics of dementia patients

Thumbnail Image
Zakipour, Seyedeh Ghazaleh
Journal Title
Journal ISSN
Volume Title
Dementia is primarily associated with memory problems in older adults, though it can also affect speaking abilities. Speech aphasia in affected patients is characterized by long pauses, unfinished sentences, and word-finding difficulties. Recently, cognitive exercises have been shown to enhance cognition in older adults and adults with dementia; though not all dementia subjects suffer from speech aphasia. Thus, it was the effort of the present research to evaluate whether cognitive exercises similarly improve speech in the dementia population in general. In this pilot study, 8 participants (4 females, age 77.13 ± 8.32) previously diagnosed with dementia were recruited from an ongoing neuro-cognitive rehabilitation clinical trial (ClinicalTrials.gov NCT03587012). The primary outcome measure was the Wechsler Memory Scale IV (WMS-IV) assessment. In two sections of the assessment, participants must recall two short stories. These spoken recalls were stored in audio files for speech analysis. As such, acoustic and linguistic features were extracted from the audio files and their transcriptions. Due to the limited sample size of the pilot study, an online dataset (DementiaBank (Pitt corpus): 166 Alzheimer’s disease, 64 healthy controls) was utilized to evaluate the extracted acoustic and linguistic features for statistically significant differences between participants with probable Alzheimer’s disease and age-matched healthy controls. In turn, significant acoustic and linguistic features were analyzed in the pilot study dataset. The results of the analysis of the collected data indicated that the linguistic abilities of dementia patients improved from baseline to post-intervention after the 4-week period of memory exercises. Moreover, linguistic features revealed more interpretable distinctions between controls and dementia patients using the online databank. In addition, the acoustic features: fundamental frequency, noise to harmonic ratio, Mel-frequency cepstral coefficients, voiced duration, harmonic differences, bandwidth of the first formant, shimmer, spectral entropy, fractal dimensions, and number of zero-crossings showed high correlations with WMS-IV scores. This pilot study indicates that cognitive trainings have benefits for the speaking abilities of dementia patients, but to confirm the results, a much larger dataset is required.
Dementia, Alzheimer’s Disease, Cognitive Exercise, Speech Analysis, Machine Learning