A plurality of key switches are disposed in a predetermined arrangement, such as a matrix arrangement, and a tone generator includes a memory for storing tone (waveform) data corresponding to the key switches. Sampling section acquires an audio signal from an external source, cuts out tone data from the acquired audio signal and writes the cut-out tone data into the memory in association with the key switches. Any one of the key switches is designated on the basis of switch operation by a user or on the basis of automatic performance information, so that, of the tone data stored in the memory, the tone data corresponding to the designated key switch is sounded, i.e. audibly reproduced.