There are strong demands to expand captioned broadcasting for TV news programs in Japan. However, keyboard entry of captioned manuscripts for news programs cannot keep pace with the speed of speech because Japanese use ideographic characters. To realize simultaneous subtitling of Japanese news programs, a real-time speech recognition system to handle broadcast news transcription has been developed. Experiments showed that word recognition error rate of the system for studio announcer speech is 2.8% in the real-time recognition and the error rate can reduce to 0.8% by manual correction with an average total delay of 10.4 seconds.
展开▼
