Research And Implementation Of MELP Vocoder Based On FPGA

Digital radio has become the forcus research of resent years due to its advantagesin spectral efficiency, call quality, reliability and confidentiality of communicationswhen comparing with anlog radio. ASIC solutions are an effective way to reducecosts, will be the development direction of digital radio. DPMR standards have madevery high voice codec requirements to digital radio in a very narrow channel forvoice signal transmission in real time and better quality. Therefore, the study oflow-rate voice codec ASIC has important application value.Firstly, in this paper the MELP algorithm is analyzed and studied in-depth, theimplementation of key decoder module is described in detail such as LPCcoefficients and LSF coefficients conversion in linear predictive analysis, multi-bandmixed excitation synthesis and speech synthesis. Secondly, in order to get a bettervoice quality in high-noise environments, we have introduced a new method whomis achieved by combining independent component analysis and spectral subtractionmethod to speech enhancement in MELP algorithm, the entire MELP vocoderalgorithm has implemented and verified on the MATLAB platform and the syntheticspeech has been tested and the qualities can meet the communication requirements.Some implementation circuits are optimized base on hardware design principles,systems principles and synchronous design principles of FPGA’s design, usingpipelined design to instead the traditional inverse discrete Fourier transform toachieve IFFT and binary representation to instead the division to save computingtime. Finally, verilog language is choosed to implement mixed excitation synthesis,LPC and LSF conversion in linear predictive analysis and adaptive spectralenhancement module, each module is simulated and verified, and the simulation andverification resluts show that the design meets the requirements of low-rate speechcoding.