To output synthesized voices of mutually different voice quality by voice input/output devices so that they can easily be discriminated from output synthesized voices of other voice input/output devices.

This synthesizer has a voice recognition part 11 which recognizes an input voice, a voice feature storage part 11z which stores features of the input voice obtained during the voice recognition, a voice data storage part 13z which stores output voice data for outputting a voice, and a voice synthesis part 13 which synthesizes an output voice and further has a voice quality change part 12 which can change the voice quality of the output voice data by using the feature information on the voice obtained during the voice recognition. Here, the voice quality change part 12 may be a voice change part 12 which automatically generates a target value of the voice quality of the final output voice and sequentially changes the voice quality of the output voice data toward the voice quality of the target value when the voice quality of the output voice data is different from the voice quality of the target value.