ETRI-Knowledge Sharing Plaform

KOREAN
논문 검색
Type SCI
Year ~ Keyword

Detail

Conference Paper Word Boundary Unconstraint Viterbi Search For Robust Speech Recognition
Cited - time in scopus Share share facebook twitter linkedin kakaostory
Authors
Hoon Chung, Sung Joo Lee, Yun Keun Lee
Issue Date
2009-10
Citation
Asia-Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC) 2009, pp.627-630
Language
English
Type
Conference Paper
Abstract
In this paper, we propose a word boundary unconstraint Viterbi algorithm for robust speech recognition against endpoint detection error on isolated word task domain. In general, onekeyword spotting framework is used to achieve robust recognition by absorbing non-speech events with acoustic filler models. One drawback of such an approach is that there is little improvement or it can even hurt performance if unexpected non-speech events occur, whose spectral characteristics are not trained into acoustic filler models. However, it is unrealistic to prepare acoustic filler models considering all kinds of non-speech events especially occurring in mobile environment. Therefore, we propose another approach where non-speech events are absorbed implicitly by relaxing endpoints constraint of Viterbi algorithm. The experimental results show that the algorithm reduces the word error rate from 80.2% to 10.6% for inaccurately endpoint-detected utterances while consuming a little more computation.
KSP Keywords
End Point Detection(EPD), Non-speech, Robust recognition, Spectral characteristics, Speech events, Viterbi Algorithm, detection error, isolated word, mobile environment, robust speech recognition, word error rate