To facilitate the generation of voice piece and to easily change the speaker of synthesized voice without requiring a large scale computer by conducting a necessary labeling process while generating voice pieces and constituting a circuit which process voice signals and generates voice information and a circuit which performs voice synthesis on one computer system.

The lyric lines of a song which is beforehand registered by a music reproducing circuit 2 are displayed and melody sound is outputted while making a user sing the song. Then, a voice information extracting circuit 3 fetches the singing voice of the use, divides the singing voice for every sound syllable based on the information such as the length of each note of the singing voice and the song as well as the tempo and puts labels for every sound syllable to generate voice pieces. When text data, which are the object of voice synthesis, are inputted, the voice pieces are combined in accordance with the data and synthesized voice is generated.