A sound generation device is composed of a main unit and a game cartridge storing a sound generation program. The game cartridge includes an XY-axes acceleration sensor for detecting a tilt in two respective directions of a game device housing. When a button of the main unit is pressed, a CPU of in the main unit reads waveform data corresponding to one syllable in lyrics from human voice sound waveform data stored in a program ROM, changes a frequency and an amplitude of the waveform data in accordance with the obtained amounts of tilts in two directions, and outputs the processed waveform data from a loudspeaker as a sound. Thus, it is possible to provide a sound generation device capable of outputting a sound by changing its pitch and volume.