Dynamics value is acquired intermittently at predetermined time periods, and a waveform data set for a sustain tone, corresponding to the acquired dynamics value, is specified from among waveform data sets stored in a memory. To generate a tone waveform while switching to the specified waveform, a waveform switching time is used which is modified suitably in accordance with a dynamics value variation amount from a predetermined time earlier than the current dynamics value acquisition time to the current dynamics value acquisition time. Such arrangements not only can variably control a tone, having a tone-color-variation realizing characteristic, in accordance with the input dynamics value but also permits a tone color variation with an enhanced responsiveness without causing the tone color variation to impart a feeing of undesired step-like unsmoothness, thereby synthesizing a tone with a high quality faithfully reproducing a desired tone color variation.