摘要
在增强型混和激励线性预测(MELPe)模型的基础上,提出了一种高音质的600 bps声码器算法。保持MELPe算法特征的同时,利用帧间参数冗余,进行多帧联合量化;运用基于预测的分级矢量量化(PMSVQ)算法对线谱频率(LSF)参数进行量化。在非正式的主观语音质量测试中,合成语音质量优于传统的LPC10e声码器,接近2 400 bpsMELP标准的合成语音。
This paper describes a new speech coder at 600 bps based on the enhanced mixed excitation linear prediction （MELPE） algorithm, which is the new NATO standard STANAG 4591. The algorithm of this speech coder includes features of MELPE. Multi - frame joint vector quantification is performed, which takes advantage of inherent inter - frame redundancy. The predicted multi - stage vector quantization （PMSVQ） algorithm is designed to quantize the line spectrum frequency （LSF） parameters. The informal subjective quality tests show that the speech quality of the proposed coder is better than that of LPC10e and approximates to 2.4 kbps MELP standard.
出处
《空军工程大学学报（自然科学版）》
CSCD
北大核心
2007年第3期70-73,共4页
Journal of Air Force Engineering University(Natural Science Edition)
基金
陕西省自然科学基金项目资助(2006F40)
关键词
语音编码
线性预测
矢量量化
speech coding
linear prediction
vector quantization
