Simultaneous synchronization of text and speech for broadcast news subtitling

Abstract

In this paper, we present our initial effort in automatic generation of subtitle for live broadcast news programs, utilizing the fact that nearly perfect transcriptions are available. Instead of using the former error-prone automatic-speech-recognition (ASR)-based method, we propose to formulate the subtitling problem as synchronization of text and speech, which is further simplified into an anchor points estimation problem. The Viterbi algorithm for hidden Markov model (HMM) is augmented with new criterions for the online anchor points estimation. Experiments indicate that our proposed methods show satisfying performance for the simultaneous subtitling application. We also present a brief introduction into our whole subtitling system under further development. © 2009 Springer Berlin Heidelberg.