论文部分内容阅读
４～８ｋｂ／ｓ变速率有限状态代数码激励线性预测语音编码（ＶＲ－ＦＳ－ＡＣＥＬ）是一种具有延时较短、合成语音质量高、算法复杂度较低的语音编码算法．在线性预测（ＬＰ）参数量化上，使用预测式分裂式矢量量化，获得很高的量化效率．在自适应码本搜索上，采用了有限状态控制分数延时搜索的算法，有效地降低了运算量．对于随机码本，采用了具有多模结构的代数码本，提高语音合成质量．对于激励码序列的增益，采用了预测式矢量量化，有效地提高了量化精度．经非正式听音测试，其中４ｋｂ／ｓ的合成语音质量超过了北美 ８ｋｂ／ｓ ＶＳＥＬＰ，接近长途质量，而６ｋｂ／ｓ和 ８ｋｂ／ｓ合成语音质量达到了长途质量，与 Ｇ．７２９８ｋｂ／ｓ ＣＳＡＣＥＬＰ相当．
4 ~ 8kb / s Variable Rate Finite State Algebraic Code Excited Linear Predictive Speech Coding (VR-FS-ACEL) is a speech coding algorithm with short delay, high synthetic speech quality and low complexity. In the prediction of linear prediction (LP) parameters, predictive split vector quantization is used to obtain high quantization efficiency. In the adaptive codebook search, the finite state control algorithm of fractional delay search is adopted, which effectively reduces the computational complexity. For random codebook, using algebraic codebook with multi-mode structure, improve the quality of speech synthesis. For the gain of the excitation code sequence, a predictive vector quantization is adopted, which effectively improves the quantization accuracy. After informal listening test, 4kb / s synthesized voice quality surpassed North American 8kb / s VSELP, approaching long-distance quality, and 6kb / s and 8kb / s synthetic voice quality achieved long-distance quality, and G. 7298kb / s CSACELP quite.