During the training phase, codebooks based on extracted features are generated via vector quantization approach 在训练阶段,每一个说话者通过矢量量化产生一个码书(语音数据库)。
Both of the optimized methods for lsp parameters and algebraic codebooks can provide the similar performance as the standard method with less search load 对分裂矢量量化码本搜索方式的优化和代数码本搜索方式的优化都在不明显降低合成语音质量的情况下有效地减少了运算量。
In order to improve the identification rate, this paper studies a kind of method that makes the compensated vq codebooks approach the codebooks trained in the testing environment by compensating the codebooks in the linear spectral domain 通过对vq码本在线性频谱域进行噪声补偿,使得补偿后的vq码本逼近测试环境下训练出的码本,从而提高说话人辨认系统的性能。
In order to improve the identification rate, this paper studies a kind of method that makes the compensated vq codebooks approach the codebooks trained in the testing environment by compensating the codebooks in the linear spectral domain 通过对vq码本在线性频谱域进行噪声补偿,使得补偿后的vq码本逼近测试环境下训练出的码本,从而提高说话人辨认系统的性能。
In order to improve the identification rate, this paper studies a kind of method that makes the compensated vq codebooks approach the codebooks trained in the testing environment by compensating the codebooks in the linear spectral domain 通过对vq码本在线性频谱域进行噪声补偿,使得补偿后的vq码本逼近测试环境下训练出的码本,从而提高说话人辨认系统的性能。