高质量4kbit/s码速率语音编码算法研究

４ｋｂｉｔ／ｓ有限状态代数码激励线性预测语音编码算法ＦＳ－ＡＣＥＬＰ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，利用了语音帧内和帧间的相关性，对线谱对（ＬＳＰ）参数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，在保证合成语音质量的同时，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，４ｋｂｉｔ／ｓＦＳ－ＡＣＥＬＰ的合成语音质量超过了北美８ｋｂｉｔ／ｓＶＳＥＬＰ，接近Ｇ．７２９８ｋｂｉｔ／ｓＣＳ－ＡＣＥＬＰ，ＭＯＳ分约为３．９。

Study on the high quality 4kbit/s speech coding algorithm

kbit/s Finite State Algebraic Code Excited Linear Prediction (FS ACELP) is a high quality speech coding algorithm which has shorter time delay and low calculation burden. For line spectrum pairs, this algorithm used a high efficient split vector quantizer based on intra frame and inter frame correlations. For adaptive codebook searching, it used a finite state fractional delay adaptive codebook which has low calculation complexity and high performence. This algorithm also used a state controled multimode algebraic codebook and a predictive vector quantizer for gains to forther improve the quality of synthesized speech. The informal tests show that the speech quality of 4kbit/s FS ACELP is much better than that of IS 54 8kbit/s VSELP and near to that of G.729 8kbit/s CS ACELP, the MOS score is about 3.9.