线谱频率及差分线谱频率参数相关性分析

Correlation analysis of LSF and differential LSF parameters

根据英语/汉语男女声线谱频率(LSF)参数及差分LSF参数帧内相关性统计结果,提出适合于LSF参数及差分LSF参数的分裂矢量量化(SVQ)分组方案。实验表明,在不考虑码书大小的情况下使用SVQ量化10阶LSF参数时,(4,6)分组的量化效果较优,否则(4,2,4)或(4,4,2)分组的量化效果较优。通过相关程度分布表清晰表明,至少68%的差分LSF参数在帧内呈微相关,有效减少了LSF参数的帧内冗余信息。随后采用DSQ和多种分组的EEDSVQ对差分LSF参数进行量化,结果表明差分LSF的量化性能优于LSF参数的量化性能。在语音编码中,采用差分LSF参数代替LSF参数作为模型参数,可在保持相同语音质量的情况下进一步降低编码速率。

In the light of intra-frame correlation of Line Spectrum Frequencies（LSFs） and Differential LSFs（DLSFs） from English/Chinese female/male speech database,an optimal partition scheme for LSFs and DLSFs was proposed.The experimental results show that if the size of codebook is not limited,dividing 10-order LSF vector into two sub-vectors of（4,6） can get better quantization performance,otherwise dividing LSF vector as（4,2,4） or（4,4,2） can get better quantization performance.The intra-frame correlation between DLSFs is significantly smaller than that between LSFs,and at least 68% of DLSFs has feeble intra-frame correlation.DLSFs were quantized by DSQ and EEDSVQ,and the experiments show that the quantization performance of DLSFs is better than that of LSFs.In speech coding systems,adopting DLSFs instead of LSFs can get less spectral distortion and reach high quality speech at lower bitrates.