39 Philipp Klumpp Computer Vision News Computer Vision News that Parkinson’s disease (PD) patients lose the ability to reliably produce speech sounds that require a high degree of muscle tension. Compared to a black box model which learned from pathological data how to estimate an articulation score between 0 and 1, an estimate of the speech sound inventory which a patient can still produce reliably carries much more information for the clinical expert and allows them to draw conclusions on their own. The proposed methods were not limited to the analysis of isolated speech sounds but could also be used to model the articulatory mobility inside the vocal tract. It is well documented that PD causes muscle rigidity, and this also affects the tongue, a complex muscular structure which plays a major role in the articulation of vowels. Philipp proposed a method to estimate the articulatory space – or, in simple terms, the maneuverability of the tongue – inside the vocal tract of an individual. It strongly correlates with the speech quality annotated by a speech and language therapist, as shown in Figure 1. The model used for the analysis of pathological speech is available here. Figure 1: Over the course of 16 recording sessions with a single PD patient, the blue line indicates the (normalized) expert rating of speech impairment (higher score reflects stronger impairment), while the orange line represents the reduction of available articulatory space.
RkJQdWJsaXNoZXIy NTc3NzU=