Design And Implementation Of Language Identification System For Web Video

Online video is one of the main information carriers on the Internet. With the rapid development of online video servise, it becomes more necessary to retrieval and supervise these videos. Traditional online video content detection technology is focused on the field of image and video, and often ignore the voice content. Since speech content is also an important part of the network video, the accurate recognition of language is the necessary condition for the further detection of speech content.The work of this paper is based on a research project from the company the author internships. The research focuses on the detection and prevention of violent terrorist video in the Internet. The author independently completed the following work:Design and implementation of language identification (LID) system. The system adopts a layered modular architecture, including pre-processing, language feature extraction, language identification and other functional modules. In the preprocessing module, the input data is processed to reduce noise, reduce the redundancy and data normalization, including the two steps, endpoint detection and acoustic features extraction. In the language feature extraction module, total variance factor (i-vector) is extrated from the video file as the langugage feature, based on factor analysis theory. Finally, support vector machine is used to train and predict languageclassifier in the language identification module.Application of language identification system. The system is applied to the identification of specific types of language, including the Turkic language subfamily, Mandarin Chinese and English. For an input video, the language identification result is submitted to the existing voilent terrorist video detection system as a detection hint. The result of the experiments shows the correlation between the Turkic language subfamily and violent terrorist video, demonstrates that the fusion of video content detection and LID may get better performance. Thus achieves the design goal.