The Design And Implementation Of A TV Speech Content Classification System

With the further promotion of triple play, the openness, integration and complexity of network also continuously increase. It provides users with rich content and high-quality experience, but, meanwhile, the influx of large amounts of information also brings about information security problems, which need urgent treatment and regulation. All kinds of media contents (text, images, audio, video) need to be placed, in accordance with relevant rules, under real-time monitoring based on auxiliary tools and support facilities, so as to detect suspicious or unqualified broadcast contents timely and make corresponding strategic decision and processing.The research object of this thesis is TV voice contents. Before identified and filtered, voice contents need to be classified first. This thesis, on the basis of the existing classification technology, proposes to design classifiers based on the combination of Rule Classification and SVM Decision Tree Classification. Different classifiers are designed according to different classification categories. After the extraction of different characteristic vectors, classification models conduct targeted sample training extract. Multiple classifiers combine and cooperate with a due division of work, and make up TV voice classification system with model training subsystem,Through constant experimentation, the selection of related parameters and functions of the system are at the same time optimized and adjusted based on the results of experiments. That reflects the system’s ability to learn. The results of experiments showed the classification system effectively improved the accuracy of TV voice classification.