Design and evaluation of a virtual reality exposure therapy system with automatic free speech interaction

Research on Virtual Reality Exposure Therapy (VRET) to treat social phobia is not new. Still, few studies focus on creating an elaborate conversation between the patient and characters in a virtual environment. This study focuses on techniques to run a semi-scripted conversation between virtual characters and a patient considering both manual and automatic speech response. Techniques evaluated are a speech detector and a speech recognizer. They were compared to a human control condition. We analysed the flow and interaction individuals (N = 24) experienced and did a Turing-like test. A case study with two phobic patients was also conducted. Both the patients and therapist, and their interaction with the system, were observed. The study showed that the different automatic techniques had their (dis)advantages, but often did not show any significant difference with the human control condition. A VRET system with semi-scripted conversations might therefore be suitable for the treatment of patients with social phobia. Using automatic speech response techniques might reduce the system workload demand placed upon therapists, allowing them to devote more attention towards monitoring the patient.