混淆网络在音频文档检索系统中的应用研究

Application research of confusion network in spoken document retrieval system

给出了一个基于音节混淆网络的语音文档内容检索系统,提出了一种基于两阶段解码的查询自动扩展方法,首先通过Viterbi解码算法在混淆音节网格上计算混淆音节的似然得分,然后利用A*解码算法从音节格上产生易混淆的扩展项,扩展项由其置信得分与阈值的比较自动产生。实验结果显示该方法能够有效提高系统的检出率。

A syllable confusion network based spoken document content retrieval system is presented.A two-stage decoding based query automatic expansion method is proposed.It is achieved by first using a modified Viterbi decoder to calculate confusion syllable likehihood score in confusable syllable lattice,and then running A＊ search algorithm to generate mostly confusable phrases from lattice.The expansion terms are automaticly generated by its confidence score.Experimental results show that the proposed method can effectively improve term detection rate.