Research On The Key Parameters Extraction Algorithm And Hardware Implementation In Speech Coding System Under Low Signal-to-noise Ratio

Low rate speech coding is an important development direction and research hotspot in the field of speech communication.One of its core problems is the extraction of speech coding parameters.Many parameter extraction algorithms can only be strictly applied in the pure voice or high signal to noise ratio(SNR)environment,however,in the low SNR environment the noise will greatly influence the accuracy of parameter extraction,which impacts on the intelligibility and quality of synthesis speech and it is difficult to guarantee the quality of voice communication.Therefore,it is meaningful to study the parameter extraction algorithm of low-rate speech coding in low-rate environment.In practical application,the speech coding algorithm is usually implemented by DSP processor.Along with the development of Internet of things,intelligent hardware and smart home device,the miniaturization of voice communication terminal becomes important and it becomes one of the major problems to reduce the hardware implementation area in speech coding implementation.Meanwhile,facing the complex application scenarios,the complexity of speech coding algorithm is also improved.The traditional DSP processor solution is difficult to solve the contradiction between the improving complexity of algorithm and lower hardware implementation area.In this case,the custom-made ASIC became the better solution.Therefore,the ASIC implementation of speech coding algorithm has higher practical significance and value.At present,the main parameters in speech coding algorithms include speech endpoint,pitch of speech and linear prediction parameters.Based on these three important parameters,oriented to the application environment of low SNR,this paper improves the parameter extraction algorithms and completes the hardware optimization and RTL code implementation.In the study of speech endpoint detection algorithm,this paper introduces the spectral entropy into traditional energy-zero-rate-ratio(EZR)method as a correction in low SNR environment and proposed the spectrum entropy weighting EZR method in low SNR environment.In the study of speech pitch detection algorithm,based on the principle of the spectrum subtraction in speech enhancement field,this paper improves the cepstrum method and proposed the improved cepstrum pitch detection algorithm combined with spectral subtraction.In the study of linear prediction algorithm,based on the geometric average lattice method,this paper proposes the covariance lattice method based on Burg algorithm and simplify the complex computation and makes it more suitable for hardware implementation.The simulation results show that the three improved parameter extraction algorithms have a better performance in low SNR environment.Hardware implementation ensures the algorithm function and operation speed,the simulation results prove that the scheme assures processing speed and the calculation results.