Question

谁能帮我翻译一下这段英文文献啊？

Vector quantization (VQ) based speaker recognition is a conventional and successful method [l]. The basic idea in this approach is to compress a large number of short-term spectral vectors into a small set of code vectors. A codebook can also be viewed as a generalization of the long-term average where the short-term spectral variations due to different textual content are not averaged out but are modeled by separate code vectors. The successful modeling of the underlying acoustic classes allows the VQ-based system to achieve high recognition accuracy even with very short test utterances. A VQ codebook is usually trained with the LBG algorithm [2] to minimize the quantization error when replacing all feature vectors with their corresponding nearest code vectors. The code-book trained based on above criterion tends to represent the density and clustering of the training data. One weakness of using the LBG codebooks as a classifier is its weak discriminative power due to the fact that only the samples within a class, but no competitive data, have been used during the training process. In other words, minimizing the quantization error of the codebook does not necessarily lead to an optimal classification performance.

谢谢大家了。
不要用google，有道吧，都不通顺啊。谁会啊 帮帮我吧 谢谢了。。

Answer 1

矢量量化（量化）的说话人识别是一个传统的和成功的方法[ 1 ]。在这种方法的基本思想是压缩短期大量光谱向量到一小部分代码载体。一个码也可以看作是一个泛化的长期平均水平在短期谱的变化，由于不同的文字内容不平均，但模型是由单独的代码载体。成功建模的基本声类允许基于矢量量化的系统达到高的识别精度甚至很短的测试话语。矢量量化码书通常是训练有素的算法[ 2 ]减少量化误差时，将所有特征向量及其相应的最近的代码载体。码本训练基于上述标准往往代表密度和聚类训练数据。一个弱点的使用群码作为分类器是其脆弱的歧视性权力由于只在类的样本，但没有竞争力的数据，已被用于在训练过程。换句话说，尽量减少量化误差的码并不必然导致一个最佳的分类性能。

Answer 2

基于发言者认识的矢量化是一种比较传统和成功的方法。这种方法的基本思想是将一系列的光谱速度压缩为一系列的编码速度。这个编码本可以看做是长期均值的一个广义，当短期光谱变化由于

Answer 3

..