A question as to the physical condition and a question as to the feeling are asked, answers are accepted by voice, acoustic features are extracted from the answer to the question as to the physical condition, and character string information is extracted from the answer to the question as to the feeling. A correlation between the acoustic features and the character string information is set, and character string information is identified from a newly accepted acoustic feature, thereby performing feeling presumption. Feeling is presumed from the voice in response to the subject's severalty and changes in the subject's age and physical condition. A medical examination by interview on the mental condition is performed on the subject by voice output and voice input, and the subject's mental condition is diagnosed based on the contents of the subject's answer and analysis of the answer by voice.