Identification of language-independent speech and language biomarkers characteristic of dementia

Dementia encompasses various progressive neurological disorders affecting cognitive functions, including language processing. Though there is no cure yet, medical research is progressing and gaining significant insights into the development of the disease. Massive screening using brain PET scans is not affordable, which motivates research into cheaper language-related biomarkers. The goal of the project is to propose new multilingual or language-independent biomarkers and decision models, supported by extensive and diverse databases as well as the automation of the feature selection and extraction. Acoustic features and fluency metrics have shown to be good biomarkers. We will investigate the potential of representations that are learned in an unsupervised manner, hence allowing to leverage massive amounts of (cheap) language data. Automated speech intelligibility and fluency measurements will be carried out to extend already validated feature sets and to give insight in speech fluency disorders related to dementia. We further investigate how multi-lingual natural language processing (NLP) models can capture dementia-related changes in vocabulary, syntax, semantic, and discourse patterns. The multilingual approach makes the results applicable in a broader geographic context and allows to pool more dementia-related data. Next, fusion of NLP and acoustic features will be researched. The results achieved throughout the project will lay the foundation of a diagnostic support tool. The schematics and a technology demonstration of the achieved results will help to move towards building larger projects with a wide range of collaboration among speech technologists, clinicians, language technologists and other entities interested in applying speech technology in healthcare programmes.

Keywords:speech processing, natural language processing, diagnostics, dementia, artificial neural networks

Disciplines:Audio and speech processing, Natural language processing, Audio and speech computing