自适应矢量量化在语音识别中的应用

介绍了用离散隐马尔可夫模型(DHMM)构造孤立词语音识别系统中的特征向量矢量量化的码书构造过程。以往的矢量量化通常采用基本算法LBG,在此基础上,引入了一种时间复杂性和空间复杂性有所降低的改进算法。该算法的核心是自适应地生成初始码书以及对初始码书的二次构造过程。从降低时间复杂度和节省存储空间,提高识别率的角度,对该算法进行了讨论。

Application of Adaptive Vector Quantization in Speech Recognition

This paper introduces the DHMM model to construct isolated word speech recognition,more important,construct code book vector quantization by the eigenvector.In former,basic LBG arithmetic of vector quantization is more common.Then,we introduce a improved new arithmetic,which reduces complexity in time and space than LBG arithmetic.The new arithmetic saves the storage and improve the recognition ratio.The core of new arithmetic creates an adaptive codebook and produces the second compressed codebook from the one.we will discuss the arithmetic from reducing the time complexity,save storage and improve the recognition ratio.