The current demonstration is designed for the medical interview domain, with the doctor speaking English and the patient speaking Arabic. At this point in the project no formal evaluation has taken place. However, informally, in office-like acoustic environments, accuracy within domain is well over 80%.
