To simply and inexpensively restore a mechanically inputted character data by diverting the character data which are mechanically inputted when document data are generated as they are and transmitting them as voice information to an opposite party.

The character data are inputted in Roman character and temporarily stored in a work processor. The saved Roman character input is then converted into a pronunciation code. When there are ≥2 kinds of pronunciation font corresponding to the pronunciation code, the pronunciation code corresponding to one optimum pronunciation font is selected in consideration of the meaning that the corresponding character data have. The selected pronunciation code is temporarily stored and sent to the opposite side (reception side). This is performed by conveying or transmitting the recording medium. The opposite party (reception side) receives and converts the pronunciation code into the digital pronunciation font corresponding to it. Further, the digital pronunciation font is converted to an analog pronunciation font and a target voice is reproduced and sent as a voice to the opposite side through a speaker 17 or headphones.