Speech evaluation of patients with Alzheimer's disease using an automatic interviewer

Abstract

The used of speech analysis in detection or evolutionary control of Alzheimer's disease and the numerous advantages that it has proven to have as screening method make that, to day, it continues raising interest in researchers. At the present day, the most recent studies are focus on automatic analysis of speech recordings rather than in the possible influence of the methodology used for obtain these recordings. The main aim of this work is to find out whether the results obtained after analyzing the recordings obtained by means of an automatic interviewer have the same statistical significance as those obtained from a human interviewer in relation to the detection or evolutionary control of Alzheimer's disease. To this effect, two methodologies for acquiring audio recordings have been compared. The first one is focuses on the use of a human interviewer to generate spontaneous speech and the second one makes use of an automatic system that, by means of visual stimuli, invites the subject to speak. For carrying out this study, Cross-Sectional Alzheimer Prognosis R2019 database has been used, in which the same speakers with AD and control have been recorded following both methodologies. A characteristics extraction process based on 5 basic temporal measures has been applied to each speech sample. Subsequently, the results obtained have been submitted to a univariate statistical analysis and a multivariate analysis in order to evaluate the discriminative capacity obtained with each methodology. The results of the statistical study of these five temporal measures used show that, regardless the methodology used in speech generation, is possible discriminate Alzheimer's patients. Furthermore, in some cases, the results obtained for the automatic interviewer have achieved performances close to those of the traditional human interviewer. These results are promising and can serve as the basis to learn about its effectiveness and extension and to further deepen the automation of interviews, especially in telemedicine and teleservice scenarios.