Medical diagnosis aided by automatic classification: Non-blackbox approaches for clinical voice assessment
06.05.2019 von 16:00 bis 17:30
Voice disorders are socially relevant, because they may lead to significant follow-up costs for health insurances and the economic system, if no adequate treatment is administered timely. Voice quality characterization is pivotal to the clinical care of voice disorders, because it aids the indication, selection, evaluation, and optimization of clinical treatment techniques, including speech therapy by administered by logopedists / speech language pathologists, and phonosurgery, performed by medical doctors specialized on voice disorders.
Current approaches to artificial intelligence, including (Deep) Neural Networks, are not fully accepted by clinical experts, partly due to their black box nature. In particular, explanatory power of these approaches is low. In contrast, we propose to use hand-crafted model based features as input to low-dimensional classification automats. Our features are meant to represent closely the properties of the voice, which are described on the level of voice production, on the level of acoustics, and on the level of perception.
Diplophonia is a particular type of pathological voice qualities, in which two simultaneous pitches are reported by clinical experts to be audible simultaneously. Diplophonia may be a symptom of a vocal dysfunction that needs medical treatment. The inherently subjective definition located on the domain of auditory perception is complemented by our approaches to track two simultaneous fundamental frequencies from high-speed videos of the vocal folds, and from audio signals. Also, first steps with a physiologically grounded hearing model are presented. The hearing model is used to predict from decomposed audio signals of the voice the presence of two simultaneously perceivable pitches.
Short bio:
ator of a research project FWF KLI722-B30 funded within the Program Clinical Research of the Austrian Science Fund (FWF), entitled “Objective differentiation of dysphonic voice quality types”. He is an organizer of the 2019 Special Session at Interspeech, entitled “Voice quality characterization for clinical voice assessment: Voice production, acoustics, and auditory perception”. He is a member of the IEEE Signal Processing Society, the Audio Engineering Society, and the Acoustical Society of America. He is reviewer for the Journal of the Acoustical Society of America, for the IEEE Transactions on Audio, Speech and Language Processing, for the Journal of Medical and Biological Engineering, for the Journal Biomedical Signal Processing and Control, and for Acta Acustica united with Acustica.