To provide a speech generator which can output speeches by changing intervals and sound volume with a simple operation.

The speech generator comprises a game system body 10 and a game cartridge 30 recorded with a speech generation program. The game cartridge 30 is packaged with an X-Y axis acceleration sensor 31 in order to detect the tilt in two directions of a game system housing 11. When an A button 16 is pressed, a CPU (Central Processing Unit) 21 built in the body 10 reads out the waveform data corresponding to one tone of lyrics from human voice waveform data 51 stored in a program ROM (Read Only Memory) 33, changes the frequency and amplitude of the waveform data according to the detected amount of the tilt in the two directions and outputs the same as speeches from a speaker 18. As a result, the speech generator can be made to sing a song.