Online Detecting End Times of Spoken Utterances for Synchronization of Live Speech and its Transcripts

摘要

In this paper, we present our initial efforts in the task ofAutomatically Synchronizing live spoken Utterances with theirTranscripts (textual contents) (ASUT). We address the prob-lem of online detecting of the end time of a spoken utterancegiven its textual content, which is one of the key problemsof the ASUT task. A frame-synchronous likelihood ratio test(FS-LRT) procedure is proposed and explored under the hiddenMarkov model (HMM) framework. The property of FS-LRT isstudies empirically. Experiments indicate that our proposed ap-proach shows satisfying performance. In addition, the proposedprocedure has been successfully applied in a subtitling systemfor live broadcast news.