Early Detection of Alzheimer's Disease Using VOT Mean Measure in New Tunisian Arabic Database

Abstract

Alzheimer”s disease (AD) is one of the neurodegenerative diseases, which first affects the function of speech in addition to a few other functions. This brings speech back to the top of promising biomarkers for the early detection of this disease before the appearance of the first preclinical symptoms. Despite the fact that there are many methods explored for early detection of Alzheimer's disease (AD), however the Voice Onset Time (VOT) values of stop consonants can be a good indicator of this disease by containing clues about the speaker's voice disorder in a language. In this study, we have proposed VOT_Mean measures to detect these deficiencies for the task of early detection of Alzheimer's disease (AD) in Arabic language using /pa/., /ta/., /ka/, and /pata/ syllables. An Arabic speech database of healthy control individuals (HC) and other with AD has been developed in association with the Alzheimer Family Assistance (AFA) center. Experimental tests were performed on this database for the didochakokinetik (DDK) task and showed that there is high significance for VOT_Mean with repect to syllables, while no significance with respect to gender. Additionnally, there is no statistical significance for the duration of words or for the duration of words in relation to sex. For the best results of the performances assessed using the VOT_Mean measurement, they were achieved with the XGBoost algorithm compared to those obtained by other machine learning algorithms for the AD and HC groups.