Volume 17, Issue S5 e049766
BIOMARKERS
Free Access

Automated speech analysis for detection of cognitive and emotional changes

Alexandra König

Alexandra König

CoBTeK (Cognition Behaviour Technology) Research Lab, University Côte d’azur, Nice, France

National Institute for Research in Computer Science and Automation (INRIA), Sophia Antipolis, France

Search for more papers by this author
Inez H.G.B. Ramakers

Inez H.G.B. Ramakers

School for Mental Health & Neuroscience, Maastricht, Netherlands

Maastricht University, Maastricht, Netherlands

Search for more papers by this author
Nicklas Linz

Corresponding Author

Nicklas Linz

Ki elements UG, Saarbrücken, Germany

Correspondence

Nicklas Linz, Ki elements UG, Saarbrücken, Germany.

Email: [email protected]

Search for more papers by this author
Radia Zeghari

Radia Zeghari

Université Côte d’Azur, Cognition Behaviour Technology Lab (CoBTeK), Nice, France

Association Innovation Alzheimer, Nice, France

Search for more papers by this author
Philippe Robert

Philippe Robert

Université Côte d’Azur, Cognition Behaviour Technology Lab (CoBTeK), Nice, France

Association Innovation Alzheimer, Nice, France

Association IA, Nice, France

Search for more papers by this author
First published: 31 December 2021

Abstract

Background

Many aspects of language and speech seem affected with increased dementia risk. Recently speech analysis has become mature enough to provide completely automatically extracted fine-grained features highly sensitive to early cognitive and affective changes.

Method

Several studies were performed during which over 150 speech samples of patients at different stages (controls, Mild Cognitive Impairment, Dementia) were collected at the clinic as well as remotely (over the phone and video-conference system). Subsamples of patients with apathy and depression were included. The spoken features extracted from the recordings were compared against data from classical assessment tools and manual annotations.

Result

Comparison to reference measures show firstly, highly accuracies in demonstrating speech differences based on clinical diagnosis (up to 90%) and secondly, significant correlations between automatically extracted and manually annotated measures (r = 0.9).

Conclusion

Speech and language features represent a promising biomarker candidate as they can be automatically and remotely extracted even over the telephone. With the help of advanced machine learning and different computational techniques the most significant markers can be identified as early indicator for screening.

1, 2

Details are in the caption following the image
 
Details are in the caption following the image