基于ANN/HMM混合模型汉语大词表连续语音识别系统  

Large Vocabulary Continuous Speech Recognition System based on Hybrid Hidden Markov Model(HMM) and Artificial Neural Network(ANN)

作　　者：蒋瑞[1] 李海峰[1] 马琳[1] 

机构地区：[1]哈尔滨工业大学计算机科学与技术学院,哈尔滨150001

出　　处：《智能计算机与应用》2012年第5期23-26,30,共5页Intelligent Computer and Applications

摘　　要：提出一种基于隐马尔可夫模型(Hidden Markov model,HMM)和人工神经网络(Artificial Neural Network,ANN)混合模型的汉语大词表连续语音识别系统。在混合模型系统中,多种模型协同工作。ANN负责建模音素发音物理特性,HMM联合语言学模型识别待识语料。这样,混合模型系统能够结合HMM和ANN两种模型的优点:HMM对时间序列结构建模能力强;ANN的非线性预测能力强,建模能力强,鲁棒性,便于硬件实现。实验结果表明,HMM/ANN混合模型系统有效结合了两种模型的优点,提高了识别率。This paper proposes a method of mandarin large vocabulary continuous speech recognition system based on hybrid Hidden Markov model（HMM） and Artificial Neural Network （ANN）. In the system, a variety of models work together. ANN models realize pho- netic properties, and HMM incorporates with linguistics model for recognizing utterances. In this way,the system can combine advantages of both ANN and HMM： HMM has high modeling ability on time sequence structure; ANN has strong ability of nonlinear prediction, model- ing and robustness, as well as facilitation for hardware implementation. The experimental results show that the system effectively takes ad- vantages of both ANN and HMM, and improves the performance.

关 键 词：大词表连续语音识别 混合模型 隐马尔科夫模型 人工神经网络模型 多路径 

分 类 号：TP393.08[自动化与计算机技术—计算机应用技术]