Design and Application of Programmable Speech Processor

為了提高通信系統(tǒng)的保密性，降低制造成本，需要進(jìn)行專用處理器的設(shè)計。該文基于SELP(Sinusoidal Excitation Linear Prediction)算法模型原理，設(shè)計了一款高質(zhì)量多速率語音專用處理器芯片。芯片使用可重構(gòu)體系結(jié)構(gòu)和超長指令字系統(tǒng)設(shè)計方法，將復(fù)雜度高的子程序進(jìn)行優(yōu)化，能夠顯著提高指令并行度。仿真結(jié)果表明：在該芯片上實(shí)現(xiàn)語音壓縮編碼算法，執(zhí)行效率高于相同工藝水平的通用數(shù)字信號處理器，并保持原有編碼質(zhì)量。該處理器能夠?qū)崿F(xiàn)多種類型的語音壓縮算法，使語音算法可以達(dá)到高保密性、低復(fù)雜度和易開發(fā)性。
關(guān)鍵詞：語音信號處理；語音壓縮；專用芯片；可重構(gòu)體系結(jié)構(gòu)；超長指令字
【Abstract】Programmable application-specific processor is necessary in order to increase communication system security and reduce manufacturecost. An application special processor is introduced for high quality multi-rate speech compression algorithm based on SELP(sinusoidal excitationlinear prediction) model. The chip uses reconfigurable architecture and VLIW(very long instruction word) system design method, by which highcomplexity sub programs are optimized to get a remarkable instructions parallel. The simulation result shows speech compression algorithms whichare implemented on this chip have higher efficiency than that on universal DSP(digitalsignal processor) under the same technical level, whilemaintaining the original coding quality. The chip can implement different types of speech compression algorithms and reach to good security, lowcomplexity and development convenience.
【Key words】Speech signal processing; Speech compression; Application-specific integrated circuit(ASIC); Reconfigurable architecture; Very longinstruction word

可重構(gòu)體系體系結(jié)構(gòu)，是指某一計算系統(tǒng)能夠利用可重用的硬件資源，根據(jù)不同的應(yīng)用需求，靈活地改變自身體系結(jié)構(gòu)，以便提供與之相匹配的體系結(jié)構(gòu)[1]。已有可重構(gòu)體系實(shí)現(xiàn)了語音算法的某些特定部分，如自適應(yīng)多速率算法中的代數(shù)碼本搜索[2]、原型編碼器中的小波變換[3]以及有限脈沖響應(yīng)濾波器[4]等，但使用協(xié)處理器使硬件規(guī)模仍然偏大，且只能對一類語音算法適用。如能在整體芯片設(shè)計中采用可重構(gòu)體系，能夠大幅度降低開發(fā)時的工作量，同時避免引入?yún)f(xié)處理器所導(dǎo)致的規(guī)模大、保密性差等缺點(diǎn)。本文基于這條技術(shù)路線，設(shè)計了一種可編程專用語音編解碼芯片TR100。