一种语音编码算法的高吞吐率实现  

High throughput implementation of a speech coding algorithm

作　　者：周瑾 薛玉群 詹毅 蒋见花 ZHOU Jin;XUE Yu-qun;ZHAN Yi;JIANG Jian-hua(University of Chinese Academy of Sciences,Beijing 100249,China;Institute of Microelectronics,Chinese Academy of Science,Beijing 100029,China)

机构地区：[1]中国科学院大学,北京100049 [2]中国科学院微电子研究所,北京100029

出　　处：《微电子学与计算机》2020年第3期9-13,共5页Microelectronics & Computer

摘　　要：本文通过对G.729语音编码算法的分析,根据实时语音处理的特性,设计一种以三帧语音数据为整体的帧轮转结构;对于算法中的相关值计算过程、滤波过程、以及硬件设计时SRAM的读取特性,结合并行结构和流水线结构,实现了语音编码算法的高吞吐量设计.完成相同计算所需时钟周期仅为优化后DSP设计的1/68.采用一组加入pink-noise噪音的语音对设计芯片进行了编解码后的PESQ音质评分测试,在信噪比不小于5 dB的情况下,语音PESQ评分均大于4.0分,完成了高吞吐率、高音质的语音实时压缩.Through thealgorithm′s analysis and the characteristic of real-time speech signal processing,a frame rotation structure have been designed.In addition,the paralleled and pipelined structures are designed for the correlation calculation,the filtering,and the read-write accessing of SRAM processes during the algorithm′s chip development.And,a storage address control logic is designed,therefore,our design′s clock cycle is only 1/68 of the DSP hardware platform operation of each frame′s speech coding.Finally,the PESQ test is completed with a set of pink-noise mixture speeches.The experimental results indicate that the PESQ are all scored above 4.0 when the SNR is not lower than 5 dB.These prove that real-time,highly compressed bit-ratio,high quality speech coding functions are successfully achieved in our design of the chip.

关 键 词：语音压缩 并行结构 ASIC G.729 

分 类 号：TN47[电子电信—微电子学与固体电子学]