4.8kbit／s语音编码技术的研究与实现

给出了一种基于改进型码激励线性预测编码（ＣＥＬＰ）算法的声码器。激励由随机码本和自适应码本构成，声道由线性预测器构成。为了获得高质量的合成语音，采用了非整数时延技术、频谱展宽技术、感知加权技术等，为了提高编码算法速度，码本搜索采用重叠码字技术，随机码本采用三元稀疏码本，最后给出了对合成语音质量的主观及客观测试结果。

Research and Realization of Speech Coding Technique at 4.8 kbit/s

We present a vocder based on improved CELP(Code-Excited Linear Predictive Coding)algorithm. The voice excitation is composed of a stochastic codebook and an adaptivecodebook,and the vocal tract is formed by the linear prediction filter. In order to improve thespeech quality,we apply some techniques such as noninteger delay , bandwidth expansion,andperceptual weighting,etc. In order to expedite the codebook search,we use the overlappingcodewords technique both in stochastic codebook search and adaptive codebook search,Thecodewords in the stochastic codebook are sparse and ternary valued. Objective and subjective test results are presented.