Parkinson disease voice features for rehabilitation therapy and screening purposes

The present work is focused on a method for the classification of voices of patients affected by the Parkinson disease (PD). In particular a binary discrimination method between PD patients and healthy subjects has been implemented. To better classify the PD patients, an additional binary method has been implemented to discriminate PD patients and non-PD patients with different voice pathologies. This project aims to produce predictive models that can be implemented in a wearable device. It processes the patient’s voice and works as a screening tool and a therapy-effectiveness estimator for both clinicians and patients. In particular, voice features such as Jitter, Shimmer, HNR and CPPS, are extracted from recorded material using both an microphone in air and a throat contact microphone. Such parameters are then processed using classification algorithms that give the best model to predict the clinical parameters used as class identifiers, such as Hoehn Yahr scale and Disease Duration (DD).