Audio Based Signal Processing and Computational Models for Early Detection and Prediction of Dementia and Mood Disorders

Neurodegenerative diseases causing dementia are known to affect a person’s speech and language. There is an increasing emphasis on earlier diagnosis of neurodegenerative disorders as evolving treatments are likely to be more effective before irreversible changes have occurred in the brain. The incorporation of novel methods based on the automatic analysis of speech signals may provide more information about a person’s ability to interact, which could contribute to the diagnostic process. This thesis demonstrates that purely acoustic features, extracted from recordings of patients’ answers to a neurologist’s questions in a specialist memory clinic can support the initial distinction between patients presenting with cognitive concerns attributable to progressive neurodegenerative disorders (ND) or Functional Memory Disorder (FMD, i.e., subjective memory concerns unassociated with objective cognitive deficits or a risk of progression). The thesis also shows that the acoustic features extracted from speech recordings for patients describing a picture can be used to construct a non-invasive and simple tool to infer early signs of dementia of Alzheimer's Disease (AD). This is further developed to firstly, identify patients with mild cognitive impairments and secondly to show a capability to assist the doctors in monitoring the progression of AD by predicting the MMSE scores in a longitudinal dataset. Further, novel acoustic features are introduced in this thesis that correlate with mood disorders such as major depression and bipolar. Combing the newly extracted features with state of the art features, led to developing a language-agnostic screening system for depression and bipolar disease. Finally, the results obtained show the discriminative power of purely acoustic approaches that could be integrated into diagnostic pathways for patients presenting with memory concerns or mood disorders. These approaches are computationally less demanding than methods focusing on linguistic elements of speech and language that require automatic speech recognition and understanding.