基于谐振结构的线性预测编码

Harmonic Structure Based Linear Prediction Coding

语音编码中一般用线性预测编码 ( LPC)建立信号的声道响应 ,如何提高 LPC的预测增益并用尽可能少的比特数量化 LPC参数将直接影响编码器的性能 .文中提出了一种基于信号谐振结构的 LPC技术 ( HLPC) ,利用 HLPC可使 LPC谱与信号谐振峰具有最佳的拟合 ,同时可获得0 .36d B的额外预测增益 .设计了 LPC参数的线性预测多层分裂码本矢量量化 ( LPMSVQ)方法 .数值实验表明 ,利用 LPMSVQ可在 2 1 bit/帧下获得透明的量化质量 .

The response of vocal tract is mostly modeled by linear prediction (LP) in speech coding.So how to improve the accuracy of analysis and quantize the model coefficients will affect the performance of the codec directly. The harmonic structure based the linear prediction coding (HLPC) was proposed. By using HLPC, the LPC spectrum can match the harmonic amplitudes more accurately and the extra prediction gain of 0.36 dB can be obtained. Also presented is a method named linear prediction multi-split vector qunatization (LPMSVQ) which is designed to quantize the LPC coefficients and the transparent quality can be obtained at 21 bit/frame.