An automatic accompanist produces tones for an accompaniment in synchronism with tones of melody produced through fingering of a human player; cue note data codes, which express tones produced in the melody, and cue time data codes, which express a lapse of time between the cue notes, are stored in a cue time track separately from an automatic accompanying track where key event data codes for the accompaniment and duration data codes each expressing a lapse of time between the key event codes; while the human player is fingering the melody, the automatic accompanist monitors the keys specified as the cue notes; if the human player does not depress the keys, the automatic accompanist stops the measurement of the lapse of time expressed by the duration data codes so as to make the accompaniment delayed.