CAJ | 학술논문

为了有效地减少语音编码的比特数、降低量化误差以及提高解码语音质量，提出了一种二级矢量量化的 LPC 声码器算法。该算法在模糊聚类与 LBG 级联的 VQ 算法的基础上，进一步采用二级矢量量化算法对特征参数矢量进行量化。特征参数为语音的两个特征值：基音周期与增益。第一级码本为矢量码本；第二级码本为误差码本。将该算法应用于 LPC 声码器中进行仿真实验，结果表明：该算法能有效地降低量化比特数并且减少了量化误差，从而使解码语音质量得到改善。
위료유효지감소어음편마적비특수、강저양화오차이급제고해마어음질량，제출료일충이급시량양화적 LPC 성마기산법。해산법재모호취류여 LBG 급련적 VQ 산법적기출상，진일보채용이급시량양화산법대특정삼수시량진행양화。특정삼수위어음적량개특정치：기음주기여증익。제일급마본위시량마본；제이급마본위오차마본。장해산법응용우 LPC 성마기중진행방진실험，결과표명：해산법능유효지강저양화비특수병차감소료양화오차，종이사해마어음질량득도개선。
In order to reduce the number of bits in the speech coding effectively ,and improve the quali-ty of decoding speech ,put forward a kind of secondary vector quantization of LPC vocoder algorithm. The algorithm ,based on fuzzy clustering and cascade LBG VQ algorithm ,further adopt two-stage VQ algo-rithm to research LPC vocoder ,codebook has two characteristic value of this parameter for voice :pitch and gain. The first level code for the training samples by fuzzy clustering and then cascade with LBG quan-titative codebook ;Secondary codebook is the first level codebook with the training sample set through the same way as the primary codebook to get the error codebook. The algorithm is used to code the LPC vo-coder in simulation experiment ,the experimental results show that the algorithm can reduce the number of bits needed for quantitative , and reduce the quantization error to improve the quality of the decoded speech.