提高连续语音识别速度的策略  

Strategies for improving continuous speech recognition

作　　者：李军[1] 刘晓明[1] 刘贤梅[1] 

机构地区：[1]大庆石油学院计算机与信息技术学院,黑龙江大庆163318

出　　处：《大庆石油学院学报》2005年第4期124-126,共3页Journal of Daqing Petroleum Institute

摘　　要：以半连续隐含马尔可夫模型HMM为例,分析了影响Viterbi算法效率的主要原因,讨论了提高该算法效率的3种实用策略,即码本剪枝、Beam剪枝及降低精度,在保证一定识别率的前提下,使得搜索处理更加有效,提高了搜索的速度.给出了3种方法的实验对比数据,实验结果表明,3种方法均可有效地提高Viterbi算法的效率.The feature of continuous speech recognition lies in the fact that it has a mass of data and complicated calculating, so improving the efficiency of recognition algorithm is important. This paper analyzes what affects the efficiency of Viterbi algorithm. Several strategies that improve the efficiency of Viterbi algorithm are discussed, including codebook pruning, beam pruning and reducing precision, which make the process of searching more efficient and improve speech of searching. Results of experiment based on these strategies are presented in the paper. The result indicates that the several methods are able to improve the efficiency of Viterbi algorithm.

关 键 词：SCHMM 语音识别 码本 剪枝 

分 类 号：TN912.34[电子电信—通信与信息系统]