Implementing machine learning techniques that can be trained to recognize vocal impressions in the lab to the real world is never an easy task. You need partners, data, validation, followed by traction and monetization. Building upon research conducted in the lab via behavioral signals processing in speech, researchers are now trying to introduce innovative techniques and implement them to real world applications. Whether this means adding a personality to virtual assistants or social robots, creating health solutions with empathy, or monitoring effectiveness of call centers, emotion recognition AI is a new field that involves deep learning and huge amounts of voice data. Consider, for example, the domain of couples counseling, screening children for autism or dealing with distressed families where crucial diagnostic and therapeutic information comes from manually-observed audiovisual data of verbal and non-verbal behavior. Behavioral Signal Processing can enable not only new possibilities for gathering data in a variety of settings – from laboratory and clinics to free living conditions – but in offering computational models to advance evidence-driven theory and practice. We will talk about the challenges, potential verticals for application and how far we are from making the perfect robot with empathy.