论文部分内容阅读
基于一个基音周期内语音线性预测残差（ＬＰＲｌｉｎｅａｒｐｒｅｄｉｃｔｉｏｎｒｅｓｉｄｕａｌ）波形，提出了一种简单而有效的ＬＰＲ谱变维矢量量化（ＶＤＶＱ）方法，即利用具有固定维数的通用码本对ＬＰＲ谱幅度形状进行矢量量化．通用码本和结构化的ＶＱ相组合减少了存储和计算复杂度，产生了高的量化效率，增强了编码语音的感性质量．计算机仿真结果表明，利用该ＶＤＶＱ技术设计的２４ｋｂ／ｓ语音编码器产生了高质量合成语音．
Based on the LPR linear predictive residuals in a pitch period, a simple and effective method of VVVV (VVVQ) is proposed, which uses the universal codebook with fixed dimension to measure the amplitude of LPR spectrum Shape vector quantization. The combination of a common codebook and a structured VQ reduces storage and computational complexity, produces high quantization efficiency and enhances the perceptual quality of encoded speech. Computer simulation results show that the 2 4kb / s speech coder designed by this VDVQ technology produces high quality synthesized speech.