Waseda Anthropomorphic Saxophonist No.5 (WAS-5) will be used in the music therapy of subjects with Spectrum Disorder (ASD). The impairment to recognize facial expression by ASD in other people causes problems when interacting with other people. WAS-5 will have a protocol to train the subjects to react with specific action to specific music then transfer that correlation to the expressions made by WAS5 through classical conditioning to differentiate expressions. This research has done a preliminary experiment with a mock program of the training program on normal people with positive effects. To effectively ensure subject fully perceive the facial expression of WAS-5, a gaze detection is also being developed using OpenCV and a USB camera. This research aims to improve the social skills of ASD subjects to better interact with other people. Further experiments with ASD subjects will be done with hardware upgrade.