基于NN/HMM混合模型的汉语地名识别系统

文章介绍了一个基于NN/HMM混合模型的汉语地名识别系统,该系统能自动判别并拒识词表之外的词。文中训练的基于HMM的模型,包括关键词模型、填充模型和“反关键词”模型。笔者对识别器的输出结果进行验证,把基于HMM的统计特征送到神经网络处理,由网络的输出来判断是否为词表之外的词。该文在实验中建立了一个基于传统N-Best方法的基准模型并试验了三种不同的网络拓扑结构,包括前馈后向传播网络、Elman后向传播网络以及可训练级联前导后向传播网络。实验结果表明前馈后向传播网络的性能最好,与基准模型比较平均错误率下降54.4%。

Hybrid Neural-Network/HMM Based Mandarin Place Name Recognition System

This paper describes the Mandarin place name recognition system with the ability of rejecting Out-Of-Vocabulary words automatically.The paper integrates neural network and Hidden Markov Models in an attempt to utilize the strength of both.HMM based acoustic models including keyword models,filler models,and an anti-keyword model are trained.Statistical features are fed to a neural network for further verification.Feed-forward backpropagation network,Elman backpropagation network,and trainable cascade-forward backpropagation network are compared.A baseline model based on conventional N-Best normalization methods is built.Experiment results show that feed-forward backpropagation network achieves the best performance,which reduces average error rate by54.4%.