摘要
差分LSF参数的动态范围小于LSF参数,可作为一种新的模型参数应用于语音编码中。分析了2种新的差分LSF参数矢量量化方法:增强差分分裂参数矢量量化(Enhanced Differential Split Vector Quantization,EDSVQ)和增强EDSVQ(Enhanced EDSVQ,EEDSVQ),并采用英语清、浊音的差分LSF参数进行分裂矢量量化实验。结果表明,EEDSVQ能有效抑制直接对差分LSF参数进行矢量量化引起的量化误差传递和叠加;在分配相同量化比特数的情况下,清音的量化效果优于浊音,为获得相同量化效果可减少对清音的量化比特数。
Differential line spectrum frequency （I_SF） parameters have less dynamic range than LSF parameters, so differential LSF can be used as new model parameters in speech coding. Two new differential LSF vector quan- tization method EDSVQ and EEDSVQ are analyzed in this paper, and split vector quantization is simulated using differential LSF parameters from English unvoiced/voiced speech database. Experimental results show that EEDSVQ can suppress the quantization error propagation caused by directly vector quantization of differential LSF parameters; the performance of unvoiced speech is better than voiced speech while allocating the same number of bits, the number of bits using for quantizing unvoiced speech can be reduced to obtain the same quantization performance.
出处
《电声技术》
2010年第11期61-64,71,共5页
Audio Engineering
