Double sides periodic substitution (DSPS) method for recovering missing speech

It has been reported that the admissible duration of missing speech is about 30 ms by previous methods; beyond this time limit, there is high distortion and poor speech quality after recovery. We have proposed a method called Double Sides Waveform Substitution (DSWS) for recovering relatively long missing speech and extended the missing time limit to 50 ms. This paper presents an improved version of the DSWS method, which we will refer to as Double Sides Periodic Substitution (DSPS). We have investigated the effects of recovering missing speech by DSPS method for 20 speakers and achieved a low distortion of 1.8 dB in 80 ms of missing speech at the frequency of 5 times per sec, and subjective hearing tests also indicated good speech quality. Further, recognition rate increased by 25% in speech recognition experiments.