Titolo della tesi: Voice Impairments in Neurologic Disorders: The Contribution of Artificial Intelligence
The human voice is a complex and multidimensional biological phenomenon which relies on the coordinated activity of integrated brain networks, in the central nervous system. Accordingly, in neurologic disorders, which are characterized by severe derangements in specific brain networks, the human voice deeply deteriorates with severe clinical and social impairments. Among technological tools useful for assessing voice abnormalities in neurologic disorders, artificial intelligence has been considered the most reliable methodology for the investigation of complex pathologic voice patterns, over the last few years. The present experimental thesis consists of a review of the most relevant studies published during the Ph.D. Program, in the field of voice analysis through artificial intelligence in neurosciences, by our group at the Department of Human Neurosciences, Sapienza University of Rome. In the studies here reported, we probed our machine-learning analysis of voice features for the automatic and objective detection of specific patterns of disease and its progression and we also objectively evaluated the symptomatic response to specific treatments, in several neurodegenerative diseases, including Parkinson’s disease, Essential Tremor and Spasmodic Dysphonia. We demonstrated that support vector machine analysis applied to voice recordings collected through smartphones during the sustained emission of a vowel was able to discriminate several groups of participants, with previously unreported high values of accuracy. Therefore, we provided the first evidence of a novel, reliable, non-invasive biomarker of disease, disease progression and therapeutic monitoring, based on advanced analysis of acoustic patterns in neurodegenerative diseases. We believe that the findings here outlined would help clinicians to design and develop novel cutting-edge approaches, based on smartphone technologies, for the assessment of patients with neurologic disorders in a telemedicine scenario. Also, novel smartphone-based recordings of human voices would improve the ecological value of our investigations since they would limit the experimental biases of laboratories. Finally, the experimental paradigm of the speech task we here proposed would allow voice recording in a culture-free scenario.