说话人识别中随机局部搜索算法的研究

目前在矢量量化的码本训练中经典的聚类方法是LBG算法,但该算法的主要缺陷是对初始码书的依赖性较大,容易过早地陷入局部极小.本文在基于矢量量化的说话人识别中研究了一种随机局部搜索的聚类算法.该算法不依赖初始条件,结构规则,容易实现,效果好,具有很优越的全局优化搜索能力,在语音参数聚类实验中表现出了很好的性能,得到的码书质量也优于经典的LBG-算法,从而为在基于矢量量化的说话人识别中设计准全局最优码书提供了一种新思路.

On a Randomized Local Search Algorithm for Speaker Recognition

The LBG algorithm is one of the common and important methods used in speaker recognition. But the main drawback of the LBG algorithm is that it often gets trapped in local optima that are significantly worse than the global optimum. This paper studies a randomized local search algorithm for vector quantization. The results indicate that the proposed algorithm is easy to implement and is competitive compared with the current best clustering methods . In addition, it is demonstrated to be more effective in the clustering for speech parameters, and can obtain better codebook quality compared with the LBG algorithm. The proposed algorithm in this paper also shows a new idea in designing the best codebook for solving more complex problems in speaker recognition.