PURPOSE: To prevent the reading error of voice data with simple constitution by storing the digital voice data read from a memory in plural memory means, comparing the contents of the memory means and controlling the output of voice data.

CONSTITUTION: Input voices are converted to digital signals and are stored into a voice memory 1. This memory 1 is read by a voice memory means 3 and is stored in the 1st memory means 6; in succession, the signal read from an access means 3 by the same address is stored in the 2nd memory means 7. The stored contents of the means 6, 7 are subjected to subtraction processing in a comparing means 12. The absence of errors is decided by a deciding means 14 when the subtracted result attains zero. The contents of the means 6 are then outputted via a voice data input means 10. The data is again read out of the means 14 and the signal 17 is supplied to the means 3 when the result of the computation is not zero. The similar reiteration is then executed. The voice memory reproducing device with which the reading error of the voice data is prevented is thus obtd. with the simple constitution.