Term-Dependent Confidence Normalisation for Out-of-Vocabulary Spoken Term Detection

作     者：Javier Tejedo Simon King Joe Frankel 

作者机构：Centre for Speech Technology ResearchUniversity of Edinburgh10 Crichton Street Human Computer Technology Laboratory (HCTLab)School of Computer Engineering and Telecommunication University Autonomous of MadridAvenue Francisco Tomás y Valiente 1128049MadridSpain 

出 版 物：《Journal of Computer Science & Technology》 (计算机科学技术学报（英文版）)

年 卷 期：2012年第27卷第2期

页      面：358-375页

核心收录：

学科分类：0711[理学-系统科学] 07[理学] 

基　　金：Engineering and Physical Sciences Research Council  EPSRC  (EP/I031022/1) 

主　　题：confidence estimation discriminative model spoken term detection speech recognition 

摘      要：An important component of a spoken term detection （STD） system involves estimating confidence measures of hypothesised detections.A potential problem of the widely used lattice-based confidence estimation,however,is that the confidence scores are treated uniformly for all search terms,regardless of how much they may differ in terms of phonetic or linguistic *** problem is particularly evident for out-of-vocabulary （OOV） terms which tend to exhibit high intra-term *** address the impact of term diversity on confidence measures,we propose in this work a term-dependent normalisation technique which compensates for term diversity in confidence *** first derive an evaluation-metric-oriented normalisation that optimises the evaluation metric by compensating for the diverse occurrence rates among terms,and then propose a linear bias compensation and a discriminative compensation to deal with the bias problem that is inherent in lattice-based confidence measurement and from which the Term Specific Threshold （TST） approach *** tested the proposed technique on speech data from the multi-party meeting domain with two state-ofthe-art STD systems based on phonemes and words *** experimental results demonstrate that the confidence normalisation approach leads to a significant performance improvement in STD,particularly for OOV terms with phonemebased systems.