| 本文已被：浏览 5858次 下载 5567次 |
码上扫一扫！ |
|
|
| 消除溢出问题的精确Baum-Welch算法 |
|
贾宾1, 朱小燕2,3, 罗予频1, 胡东成1
|
|
1.清华大学自动化系,北京,100084;2.智能技术与系统国家重点实验室,北京,100084;3.清华大学计算机科学与技术系,北京,100084
|
|
| 摘要: |
| Baum-Welch算法是在语音领域中用于HMM(hidden Markov model)模型参数训练的最基本方法之一.但它在多样本训练时存在着严重的上、下溢问题,需要不断地人工介入来调整中间参数.该文提出了一种新的能消除上、下溢问题的Baum-Welch改进算法.该算法不但摆脱了人工介入,保证了计算的精度,而且不会带来过大的计算和存储要求.实验结果表明了这种新算法的有效性. |
| 关键词: 隐马尔可夫模型,Baum-Welch算法,溢出,语音识别. |
| DOI： |
| 分类号: |
| 基金项目:本文研究得到国家自然科学基金(No.69982005)资助. |
|
| Accurate Baum-Welch Algorithm Free from Overflow |
|
JIA Bin,ZHU Xiao-yan,LUO Yu-pin,HU Dong-cheng
|
| Abstract: |
| Baum-Welch algorithm, which is often troubled with overflow, is one of the basic methods in the field of speech signal processing. People have to adjust the inner parameters constantly. So in this paper, a modified Baum-Welch algorithm is presented to avoid the overflow completely. With this algorithm manual adjustment is not needed and the computation accuracy is guaranteed. There is no significant extra cost of computation and storage. The feasibility of the new algorithm is shown in the experiment. |
| Key words: HMM (hidden Markov model), Baum-Welch algorithm, overflow, speech recognition. |
