隐马尔可夫模型(HMM,Hidden Markov Model)是语音识别中广泛采用的鲁棒性统计方法。本文采用禁止搜索(TS,Tabu Search)算法训练HMM参数,提出了基于禁止搜索的隐马尔可夫模型(TS-HMM)算法。该算法可以使搜索最优模型参数的过程达到全局优化。仿真结果表明与传统的前向-后向算法相比,TS-HMM算法具有更好的性能,且能够达到全局优化。 Hidden Markov Model (HMM) is a robust statistical method widely used in speech recognition. In this paper, we use the TS (Tabu Search) algorithm to train HMM parameters and propose a hidden-Markov model (TS-HMM) algorithm based on barring search. The algorithm can make the process of searching the optimal model parameters achieve global optimization. The simulation results show that the TS-HMM algorithm has better performance and can achieve global optimization than the traditional forward-backward algorithm.