1.8 kb/s多带线性预测编码算法设计

Multi-Band Linear Prediction Coding (MBLPC) of Speech at 1.8 kb/s

设计了一种多带线性预测编码 ( MBLPC)语音压缩算法 ,该算法结合正弦编码和线性预测编码的优点 ,对语音信号的包络谱用线性预测编码方法进行估计 ,而对激励谱用多带方式进行处理 .改进了特征参数的提取和量化方法 ,包括基于谐振结构的基音检测算法、基于谐振结构的线性预测分析以及包络谱参数的线性预测多层分裂码本矢量量化 .非正式主观测试表明 ,在 1 .8kb/s编码速率下 ,由 MBLPC算法所重建语音的平均意见分为 3.0 ,而清晰度指标约为 91 % .

A new speech coding algorithm named multi-band linear prediction coding (MBLPC) was presented. By exploiting the advantages of both sinusoidal coding (SC) and linear prediction coding (LPC), the MBLPC divides the speech spectrum into envelope and excitation spectrum, and processes them respectively. Also new methods for pitch detection, spectrum parameter estimation and quantization were introduced. The informal subjective listening tests show that the speech reconstructed by MBLPC at 1.8 kb/s has the mean opinion score (MOS) 3.0 and articulation 91%.