论文部分内容阅读
人机语声通信的重要方面之一是语声应答系统。线性预测编码(LPC)声码器技术被广泛应用在该系统以及各种语音合成器中。它能够使语音数字信号的比特率从64kb/s降至2.4kb/s,并具有较好的合成声音。但是这种LPC声码器最大的缺点是合成声自然感较缺乏。在2.4kb/s LPC声码器中,自相关格型滤波器被用来估计语音的声道参数。本文把Friedlander的前置窗格型算法应用到声码器中,提出了一种新的LPC声码器:前置窗LPC声码器。该声码器中的语音分析器是用前置窗格型滤波器对语音信号进行块处理;综合器是由本文导出的前置窗格型逆滤波器构成。该方法通过抽取一组偏向关系数和辅助参数,提高声道参数估计值的精确性,因此在比特率和计算量略有增加的代价下,可获得声码器性能的改善。计算机模拟试验结果可以得到初步的结论:前置窗LPC声码器的合成声音质量比2.4kb/s“自相关”LPC声码器有一定的改善。
One of the important aspects of human-machine voice communication is the voice response system. Linear Predictive Coding (LPC) vocoder technology is widely used in this system as well as various speech synthesizers. It can reduce the bit rate of the pronunciation digital signal from 64kb / s to 2.4kb / s, and has better synthesized sound. However, the biggest disadvantage of this LPC vocoder is the natural lack of synthetic sound. In the 2.4kb / s LPC vocoder, the autocorrelation lattice filter is used to estimate the speech channel parameters. In this paper, Friedlander’s front pane algorithm is applied to the vocoder, and a new LPC vocoder is proposed: the front window LPC vocoder. The voice analyzer in the vocoder uses a pre-cell filter to process the voice signal; the synthesizer is composed of the pre-cell inverse filter derived in this paper. The proposed method improves the vocoder performance by extracting a set of biased correlation coefficients and auxiliary parameters to improve the accuracy of the channel parameter estimation. Therefore, the performance of the vocoder can be improved at the expense of a bit increase in bit rate and computational complexity. Computer simulation results can be initially concluded: the synthesis of the front window LPC vocoder sound quality 2.4kb / s “autocorrelation” LPC vocoder has a certain improvement.