Abstract: A conflict de-escalation training system includes a reproducer of sound and images and at least one device adapted to detect speech, proximity, and movement of a human trainee relative to the reproducer. Database(s) store trigger event definitions and natural language understanding (NLU) rules. A trigger analyzer determines when one of the trigger event definitions is satisfied by trainee’s speech/actions. An intent analyzer ascertains intent of the trainee’s speech using the NLU rules. A compliancy state generator determines a state value based on at least one of the intent of the trainee’s speech, proximity, and movement. The state value indicates a level of compliance of an imaginary individual being subjected to the trainee’s speech/actions. An output generator generates facial images and speech responses for the imaginary individual based on the state value. The facial images and speech responses are provided to the reproducer for output thereby.