Evaluation of double sided periodic substitution (DSPS) method for recovering missing speech in packet voice communications

The author proposes a DSWS (double sides waveform substitution) method and a DSPS (double sides periodic substitution) for recovering missing speech. Recovering missing speech is an important problem in packet voice communications, since missing speech is a major impairment in packet network communications. The author investigates all estimation techniques for packets containing 16 ms of speech and takes comparison tests of each technique by a subjective test (opinion test) and an objective test (LPC spectral distortion). These tests indicated that the most effective estimation is the DSPS method, which extends the acceptable ratio of missing packets of 10% by the previous method, to 20% in the experimental results.<<ETX>>