The invention discloses a character dialogue subtitle extraction method oriented to a news video. The method includes the steps that firstly, a news program is converted into a digital news video through a digital acquisition device; secondly, shot boundary detection and voice preprocessing are conducted on the news video; thirdly, host detection and clustering are conducted based on voice shot sets acquired in the second step, and then shots in the same class are made to all correspond to a same host; fourthly, based on voice of different hosts acquired through a clustering result acquired in the third step, corresponding host acoustic models are built, and all voice shots in the video are divided into the host speaking shots and the character dialogue shots; fifthly, subtitle detection and tracking are conducted on the character dialogue shots, and the appearance time, disappearance time and the space position of each subtitle are acquired; sixthly, based on subtitle position analysis, background regions and noise subtitles misjudged as subtitles are filtered out, and the remaining subtitles are used as character dialogue subtitles for output.