一种识别说话者的新方法

在能够自动识别视频中的说话者的系统中，大部分采用的是声音和唇部运动相结合的方法。文中则采用了另一种方法有效地达到了目的，即通过检测人体头部和手部的运动来鉴别说话者。基于演讲者在说话时通常会伴有头部运动或是手部运动，该方法既能实现说话者的检测，又能避免由于观测点过远而导致无法判断人唇部运动的局限性。在系统的实施过程中，运用了多种图像处理方法，并且对三帧差运动法做出了改善，使其能更高效、更准确地检测到头部和手部的运动。经过多个不同的视频测试后，本系统的F1 score高达91．91％，从而验证了该系统的可行性。

A New Method of Identifying A Speaker

In systems that are able to detect speakers in a video automatically, a method based on the voice and lip move- merit is proposed generally. But in this paper, another technique, which is against detecting head movement and hand movement, is employed. Since those two kinds of body movements are always accompanied when a person is speaking, this method effectively detects a speaker and avoids the limitation of accurately identifying a lip movement because of a far ob- serving distance. During the process of implementation, a combination of several different image processing algorithms is a- dopted, and the three -frame difference algorithm is modified in order to reach a further effectiveness. The average of Fl score of this system is 9t. 91% after testing several different videos, which verifies the feasibility and accuracy.