The added value of speech technology in clinical care of patients with dysarthria

Adults with dysarthria frequently face social, psychological, and daily communication problems, negatively impacting their well-being. Therefore, they should have access to an early diagnosis and therapy. Through decades, acoustic data has potentially been used for objective evaluations of speech disorders, providing scientific and engineering insight. Considering the adverse effects of disordered articulation and prosody on communication and quality of life, the overall aim of this thesis is to contribute to the clinical care and well-being of patients with dysarthria through the development of signal-processing techniques and speech technology in general. This research increases the existing knowledge about behavioural treatments of articulation in this population and extends the understanding of prosodic mechanisms and strategies through emerging techniques which include acoustically based methods. Concerning articulation, a clinically valuable set of exercises has been developed for direct and intensive therapy according to the principles of motor learning for speech rehabilitation. Also, a comprehensive and hierarchically structured training method for articulation has been designed. The intensive articulation therapy based on this methodology has proven to be feasible and effective for improving speech intelligibility in subjects with chronic or progressive dysarthria. The methodology and exercise material were incorporated into a Virtual Articulation Therapist (VAT), which also includes visual and auditory cues of the target utterance and provides adequate and relevant feedback to end users. The VAT offers clear applicability in the medical management of patients with dysarthria, is well accepted by end-users and has an added value for the current Flemish healthcare situation. Concerning prosody, a new acoustically-based multidimensional approach has been developed for the objective assessment and more comprehensive characterization of sentence accent in atypical prosody of speakers with dysarthria. The clinically usable set of newly detected acoustic features correlating with accent, combines different measures of F0, intensity and duration within an utterance. The developed automatic algorithm can accurately predict accented and unaccented syllables in healthy and dysarthric speech of Germanic languages like English and Dutch comparable in accentuation patterns. A cross-population study validates the detailed set of acoustic descriptors of accent production, which shows high reliability. In addition, it provides an objective description of the strategies used for accentuation across different severities of the speech impairment. The introduced speech technology provides the speech therapist with an objective and automatic measure of sentence accent, supporting the understanding of atypical prosody production and helping make informed clinical decisions and select more tailored therapy approaches