Truncate processing is performed so that a new tone is not generated before the envelope waveform is finished but the new tone is produced after the volume of the previous tone is sufficiently diminished, by which the new tone can be produced with a satisfactory feeling of attack. Further, the envelope waveform is not abruptly interrupted and cleared in a moment but is decayed at a constant and high speed, thereby preventing the generation of noise such as a click. Moreover, demanded speed data is provided for each tone, by which it is possible to produce repective tones not at a demanded speed like the least common multiple but at an optimum speed for each tone and to make the previous tone generation proceed to the next tone generation without producing noise such as a click.