摘要
在中速率下,多脉冲激励线性预测编码器的合成语音质量受到参数量化方式及量化精度的很大影响。本文提出了两种有效且易于实现的新方法,分别用于改善反射系数和激励脉冲幅度的量化。初步的实验结果表明,这两种改进算法的效果相当好,段信噪比的提高在2 dB 左右,语音质量非常接近无限量化精度(即不进行量化)时的语音质量。
Multipulse excited LPC is a very promising algorithm for its good speech quality. However,the quality of synthesized speech may be degraded by low efficient quantiza- tion method.In this paper,we propose two effective methods for improving the quan- tization of PARCOR coefficients and pulse amplitudes.Experimental result shows that our methods are very successful.A total increase of about 2dB in segment SNR is achieved.Subjectively we can hardly tell the difference between synthesized speech with quantization and the one without it.
关键词
多脉冲激励
线性预测编码
参数量化
quantization
digital signal processing
linear predictive coding
multipulse excitation
