In the recording/reproducing of digital audio signals, errors are detected and corrected by using two parity words, one arranged at the center of the block formed of data words and the arranged at one end of the block. The probability that uncorrectable error will be present in the center of the block is relatively high, so placing the parity word there prevents loss of the more valuable data. Maximum correctable burst errors are determined by the length of the block, so placing the other parity word on the end of the block lengthens it and improves burst error correction. The parity words are arranged as indicated before adding a cyclic redundancy check (CRC) code to the data signal and then modulation coding the signal before recording. During playback, the reproduced signal is demodulated and the CRC code used to detect errors for which error pointers are generated. The reproduced data is read into memories in accordance with generated addresses and the pointers prevent the writing in of words found to be in error, which error words are subsequently corrected if possible using parity codes originally encoded into the signals prior to recording.