低速率语音编码器的性能分析

The Performance Analysis of Speech Coding at Low Bit Rates

本文对码激励线性预测编码器(CELPC)进行了分析,针对CELPC不能很好地重构语音高频部分以及缺乏迅速跟踪语音清、浊音转换部分的能力的问题,提出了随机码本主轴脉冲法和采用不同加权系数、附加低通滤波的自适应码本的搜索过程,以进一步提高合成语音的质量。同时,提出了一种简化的自适应码本和随机码本的联合优化法。计算机模拟结果表明,采用上述各种方法后,使CELPC的性能较原始算法有不同程度的提高。

In this paper we analyze every part of Code-Excited Linear Predictive Coder (CELP) in detail. Because the reproduction of both higher frequencies and rapid changes in speech signal is inadequate in CELP, we introduce different noiss-shaping factor for the adaptive codebook search, low pass filter of pitch excitation and stochastic codebook vector's principal axis extracting vector excitation coding to improve the synthesis speech quality. In addition, we introduce joint optimization with reduced computational load. The result of analysis and simulation show that various efforts introduced above improve the CELP performance in different degree.