The challenge of recognizing emotions from speech is more pronounced in patients with neurological disorders, such as Parkinson’s disease (PD). Apart from the fact that PD patients experience deficits in the production of emotion, other factors of the disease, such as speech impairment, affect significantly the emotional content of their speech. Our work introduces a machine learning pipeline for developing a speech emotion recognition system as a digital biomarker for neurological disorders. In this poster, we present our transfer learning pipeline: build a speech emotion recognition model by combining various publicly available emotional speech databases and use this model to infer the emotion of PD patients. Our research outcome paves the way to understand better the emotional characteristics of PD speech, which in turn will help to improve PD telemonitoring systems.