Although NHK has been subtitling broadcast news simultaneously by using a real-time speech recognition system, it is difficult to use the system for other live programs such as sports due to problems caused by background noise, speaking style, etc.. Therefore we decided to use a special speaker called "re-speaker", who listens to the speech in the program and re-speaks the content of that speech into the speech recognizer, and as a result, we realized to subtitle live broadcasts in the live programs. In this paper, we describe an outline of the speech recognition system that was actually used for live subtitling in Salt Lake City Olympic Games and effects of the re-speak method with the speech recognition experimental results and test-set perplexities.
展开▼