论文部分内容阅读
多脉冲激励■型的■为合成高质量的语音提供了新的途径。从原理上讲,当脉冲个数逐渐增多时,多脉冲串可以很好地逼近残差信号,从而合成出高质量的语音信号来,但与此同时,传输或存储所需的数据率也增加了许多。因此,如何在保持合成音质不变的前提下,尽可能地降低数据率便成为这一模型的一个重要研究课题。以此为目标,利用语音信号的非平稳特性,本文提出了变个数多脉冲激励线性预测模型(VN-MPLPC),并相应详细给出了一种基于Choleskg分解的多脉冲求解算法。文章最后介绍了一个采用VN-MPLPC和VQ技木的语音分析——合成系统,反复的实验结果说明了新方法的有效性。
Multi-pulse excitation ■ ■ ■ for the synthesis of high-quality voice provides a new way. In principle, when the number of pulses is gradually increased, the multi-pulse train can approximate the residual signal well to produce a high quality speech signal, but at the same time, the data rate required for transmission or storage also increases A lot. Therefore, how to reduce the data rate as much as possible under the premise of keeping the synthesized sound quality unchanged becomes an important research topic of this model. Taking this as a target, this paper proposes a variable-number multi-pulse excited linear prediction model (VN-MPLPC) by using the non-stationary characteristics of speech signals. A multi-pulse solving algorithm based on Choleskg decomposition is given in detail. Finally, the article introduces a speech analysis - synthesis system using VN-MPLPC and VQ technology, and repeated experimental results illustrate the effectiveness of the new method.