Automatic broadcast news transcription converts speech into text by a large vocabulary continuous speech recognizer (LVCSR).This technique is an important prerequisite to various tasks

e.g.

structural segmentation

semantic access and content-based retrieval of broadcast news.In this paper

we develop an automatic caption generator (ACG) for Mandarin broadcast news.The system integrates various functions

i.e.

audio extraction from video

audio type classification and segmentation

speaker recognition

LVCSR

caption generation and video control.Experiments show that the system can achieve high speech recognition accuracy.A potential deployment of ACG is to help the hearing impaired and elderly people in enjoying TV programs.