Research On The Waveform Domain Anti Packet Loss Technology In Speech Communications

Although all kinds of communication systems have been developed for many years, the speech communication, as one of the basic and traditional form of communications, is still facing many problems that have not been solved effectively. Packet loss is one of the main problems that have big impact on speech quality; therefore, the research on speech anti packet loss technology is very important.In this thesis, the speech waveform domain anti packet loss technology has been studied, many modified schemes on PLC algorithms have been designed, in order to improve the quality of speech communications more effectively. The modified PLC schemes of G.711 iplc, one-sided and double-sided waveform replication, pattern matching, time scale modification algorithms have been programmed by C and Matlab languages. Some programs based on these algorithms exceed the performance of G.711iplc’s with some satisfactory results. The main work of this paper can be summarized as follows:Firstly, the overall speech anti packet loss schemes have been designed, with focus on the overall scheme of the waveform domain anti packet loss.Secondly, the waveform substitution PLC algorithms have been studied. Some in-depth research have been made on one-sided and double-sided waveform replication and pattern matching PLC algorithms. The improvements of one-sided and double-sided waveform replication include introducing of 20% linear amplitude attenuation per frame for continuous loss due to avoiding serious echo effect, using a triangular window to overlap and add for smooth transition in frame boundaries. The improvements of pattern matching include introducing of 20% linear attenuation per frame for continuous loss, using a triangular window to perform overlap and add process for smooth transition in frame boundaries, replacing correlation with normalized correlation, introducing of the short-time zero crossing rate detection after pattern matching in order to search pattern accurately.Thirdly, the pitch waveform replication PLC algorithms have been studied. Some in-depth analyzing and improvements have been applied to the G.711 iplc algorithm,including the simplification of program process, the modification of key parameters, changing methods for continuous dropped frames and pitch buffer update mode, replacing cross-correlation function with AMDF and introducing of the short-time zero crossing rate detection to extract the refinement of pitch period.Fourthly, the time scale modification PLC algorithms have been studied. A variety of PLC schemes based on the time scale modification algorithms of WSOLA and RTISI have been designed, with the aim of concealing the loss regions by stretching the speech waveforms. In addition, the RTISI PLC programs have been modified, including the modification of the scaled window coefficients, introducing of random phase perturbation in phase estimation process, in order to reduce some resonance and echo.Finally, the anti packet loss performances of various classical improved PLC schemes have been evaluated and compared by a large number of experimental data.The test results have shown that, in 1%~8% low packet loss rate and 10%~50% high packet loss rate intervals, when a single frame loss often occurs, the PLC performance of modified G.711 iplc scheme is better; in 1%~8% low packet loss rate interval, when the continuous frame loss often occurs, the PLC performance of modified DSWR with linear attenuation is better; in 10%~50% high packet loss rate interval, when the continuous frame loss often occurs, the PLC performances of modified TSM and DSWR with linear attenuation are turning better. In addition, the PLC performance of modified pattern matching is more balanced in each packet loss rate.