13 Março 2018, 11:38 - Fátima Sampaio
Candidate: Anna Maria Pompili N.º 72420
Title: Speech and language technologies applied to diagnosis and therapy of brain diseases
Location: Room 0.19, Pavilhão de Informática II, Alameda
Advisors: Prof. Alberto Abad Gareta / Prof. ª Maria Isabel Segurado Pavão Martins Catarino Petiz
Abstract: In the last few years we have witnessed a deep involvement of speech and natural language processing techniques in our daily routine. To mention some examples, voice search and voice-to-text are available to everyone through modern smarthpones, smart speakers allow to control home automation devices through voice commands, and, in the near future, we will interact with vehicles through speech. Among the many areas in which speech and language technologies can provide important benefits, health is one of the most challenging and appealing. Brain diseases, and more specifically neurodegenerative disorders, include a range of conditions that affect the brain causing an irreversible and progressive damage. There is no cure for these kind of diseases, but the early detection of the symptoms onset may mitigate their progress. The initial screening is performed through the patient’s natural history and through neuropsychological tests targeting the evaluation of specific cognitive functions, such as memory, planning, and language. For these reasons, speech becomes the primary mean to transmit information about themselves and to perform the clinical evaluation. In this thesis proposal, we investigate how current speech and language technologies may contribute to the diagnosis of neurodegenerative diseases in order to ease their screening process and to provide clinicians with an objective, additional diagnosis. To this end, we first overview the most common neurodegenerative disorders and the way they are diagnosed, then we focus on current approaches and limitations of existing speech technology solutions applied to the diagnosis of neurodegenerative dis- eases. Depending on the damage that the neurological injuries may have caused, we identified three main topics of interest. First, monitoring of cognitive functions through the administration of neuropsychological tests. In this case, the capability of producing intelligible speech is preserved, while other cognitive abilities like attention or memory may have resulted impaired. Second, analysis of discourse production through stimuli eliciting continuous speech. In this case, the injuries affect directly the areas of the brain dedicated to the processing of language. Third, assessment of motor speech disorders through specific vocal tasks. In this case, the injuries affect the organs responsible for the production of sounds causing uncontrolled movements and thus, incorrect and unintelligible speech. According to the identified areas, this thesis proposal makes the following contributions: i) on-line implementation of a representative set of neuropsychological tests used in the screening of dementia exploiting automatic speech recognition technology, ii) analysis of the topic coherence of spontaneous speech elicited with a specific test through methods borrowed from the area of natural language processing, iii) research on sensible features and tailored techniques capable of modeling diseases affecting speech production at the level of the phonatory, respiratory and articulatory systems.