Song data indicating each sound composing a song and including a step that indicates a training mode (self-running mode or wait mode) for each sound is read. When the read song data is in self-running mode, the reading of the song data is continued, regardless of whether the key to which a user is guided by the illumination of the LED is pressed. When the read song data is in wait mode, the song process is temporarily stopped at a sounding timing of the key to which the user is guided by the illumination of the LED. The song data of the next sound is read upon the pressing of the key to which the user is guided. Thus, for example, if a self-running mode is set for the sounds which can be operated easily by a key during a song, the training mode can be set for the sounds which are difficult to be operated, so as to perform particular musical performance training.