Recovery ofLSPCoefficients inVoIPSystems usingEvolving Takagi-Sugeno FuzzyModels
Eric E. Jones, Plamen Parvanov Angelov, Costas S. Xydeas · 2006
Inordertodeliver realtime, highquality voice services inpacket basedvoice system(e.g. voice overInternet protocol, VoIP)system designers musttackle inherent quality problemsrelated topossible packetloss. To combatthe inevitable speech quality deterioration resulting fromtheloss of transmitted packets ofspeechinformation, techniques that provide estimates ofthelost information thatisneededbythe speechrecoveryprocessare of considerable interest. Furthermore, in futureVoIP systemsemploying linear predictive coding(LPC)basedspeechcoders, a significant percentage ofthecodedspeech information willrepresent the valuesofLPC coefficients andthusa new approach for estimating missing LPC filter coefficients ispresented inthis paper.Thisapproach employs a new formulation ofLSP recovery systemarchitecture whereevolving fuzzyrule-based modelsand particularly so-called evolving Takagi-Sugeno modelsaredeployed togenerate therequired estimates of missing LSPs.Theproposed missing parameters estimation technique isgeneric andinitial experimental results demonstrate itsconsiderable potential inimproving thequality ofLPCbased decoded speech inVoIPapplications.