基音周期矢量量化中权重系数的计算  

Computation of weight coefficients for pitch vector quantization

作　　者：唐昆[1] 李晔[1] 徐敬德[1] 崔慧娟[1] 

机构地区：[1]清华大学电子工程系,微波与数字通信国家重点实验室,北京100084

出　　处：《清华大学学报（自然科学版）》2010年第4期569-571,共3页Journal of Tsinghua University(Science and Technology)

基　　金：国家自然科学基金资助项目(60572081)

摘　　要：为了更好地区分正弦激励线性预测声码器中待量化基音周期参数矢量各维基音周期参数的重要性,提高合成语音质量,提出了一种新的权重系数计算方法。发现权重系数的计算应考虑当前语音帧所有子带的清浊音度参数,而不仅仅是第一个子带的清浊音度参数,这样能够更好地区分各语音帧基音周期参数的重要性。将各语音帧的所有子带清浊音度参数相加,其和作为基音周期参数量化时的权重系数。测试结果表明,与仅将第一个子带的清浊音度参数作为权重系数相比,新方案能使合成语音质量MOS有较大提高。A method for computing weight coefficients was developed to better distinguish the importance of each pitch parameter in the quantized pitch vector for sinusoidally excited linear prediction （SELP） model and to improve the synthesized speech quality. The computation of weight coefficients should depend on all band pass voicing coefficients besides the first one,which better distinguishes the importance of each pitch parameter. All band pass voicing coefficients of speech frames were added to use the sum as the weight coefficient for pitch vector quantization. Tests show that the mean opinion score （MOS） obviously increases compared with the conventional strategy which only uses the first band pass voicing coefficient as the weight coefficient.

关 键 词：语音信号处理 基音周期 加权系数 语音合成 

分 类 号：TN912.3[电子电信—通信与信息系统]