A technique of enabling a singer to auditorily recognize how to change the way of singing is provided. The CPU (11) of a karaoke device (1) associates model voice data stored in a model voice data storage area (14c) with inputted learner voice data in a time axis direction. Then the CPU (11) shifts the pitch of the learner voice data so that it may coincide with the corresponding pitch of the model voice data according to the result of the association, compresses or extends the section (mora) of the learner voice data in the time axis direction so that the section length of the learner voice data coincides with the corresponding section length of the model voice data, and outputs the resultant learner voice data to a voice processing section (18). The voice processing section (18) converts the learner voice data supplied from the CPU (11) into an analog signal and generates the sound from a loudspeaker (19).