PURPOSE: To improve the working efficiency, by recording reading voice of a printed original by many speakers on everyone of many magnetic tapes, reproducing the tape for voice recognition, and performing photocomposition by means of an electronic computer by the data.

CONSTITUTION: A voice analysis circuit 1 detects wow/flutter produced due to the rotating system of a recording tape recorder 6 and a reproduction tape recorder 5 with a reference signal recorded on a specified track, and the wow/ flutter component of a signal recorded on the track based on it is eliminated at a wow/flutter eliminating circuit 11. The frequency characteristics specific to the recorder 6 are standardized by using a frequency characteristic correcting circuit 12. The voice signal thus compensated is analyzed for the frequency characteristics at the circuit 1, and the data to each voice is inputted to a voice/ character converting circuit 2. The registrated data of the speakers are corrected at the circuit 2, the frequency data are compared with the corrected registrated data, and when they are coincident with each other, characters corresponding to the voice signals are outputted.