变个数多脉冲激下低比特率高质量LPC语音合成器研究

Research on Variable-Number MPLPC Synthesizer for Producing High Quality Speech at Low Bitrate

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。

The introduction of multipulse excitation provides a new way for producing high quality apecch. Theoretically, the multipulse sequence can approach the prediction error satifactorily when the number of pulses reaches a certain value. However, the bitrate required in transmission or storage is increased intolerably at the same time, This paper focuses on how to lower the bitrate while maintaining the speech quality.By exploiting the nonstationary characteristic of speech signal, we present a variable-number multipulse excited LPC model (VN-MPLPC) . An efficient algorithm based on Cholesky decomposition is included to compute the location and amplitude of the multipulse, In the end, we introduce a VN-MPLPC speech analysis-synthesis system. Experiements have shown the efffctiveness of the new model.