论文部分内容阅读
隐Ｍａｒｋｏｖ模型（离散ＨＭＭ）的参数估计问题，是ＨＭＭ在语音处理应用中的关键问题。经典的Ｂａｕｍ＿Ｗｅｌｃｈ算法是基于最陡梯度下降的局部优化算法，ＨＭ Ｍ模型的质量取决于初始模型的设计。解决这一问题的根本方法在于使算法具有随机性。本文结合随机松弛算法（ＳＲ）的全局搜索能力和Ｂａｕｍ＿Ｗｅｌｃｈ算法的局部优化性能，提出了一种离散隐 Ｍａｒｋｏｖ模型参数的全局优化算法。该算法根据 ＨＭＭ的参数对 Ｐ（Ｏ／λ）的不同影响，对观察值概率矩阵Ｂ进行满足一定降温规范的随机扰动，可对离散ＨＭＭ的参数进行全局优化训练。
Hidden Markov model (Discrete HMM) parameter estimation problem is the key issue of HMM in speech processing applications. The classical Baum_Welch algorithm is a local optimization algorithm based on the steepest descent gradient. The quality of the HMM model depends on the design of the initial model. The fundamental way to solve this problem is to make the algorithm random. In this paper, a global optimization algorithm for discrete Hidden Markov model parameters is proposed based on the global search ability of random relaxation algorithm (SR) and the local optimization performance of Baum_Welch algorithm. According to the different influence of HMM parameters on P (O / λ), the proposed algorithm optimizes the parameters of discrete HMM by randomly perturbing the observed probability matrix B to some extent.