A Keyword Spotting Based Sports Type Determination System

This paper proposes a novel system to automatically determine the sports type of a sports game by conducting keywords spotting on short fragments (around 10 minutes) of a sports game. In this system, we first develop an audio segmentation module as a front-end to separate announcers´ speech efficiently from the complex sports audio stream. Then we employ speech recognition technology on these speech segments to extract keywords as the features of each kind of sports. Finally, based on the KWS (keyword spotting) results and the specific keywords we defined for each kind of sports, the classification is conducted based on a score ranking strategy. In order to improve the classification accuracy, acoustic model adaptation and language model adaptation are performed to improve the KWS results. MAP (maximum a posterior) adaptation is employed for acoustic model and a keyword-frequency-based adaptation method is proposed for the language model adaptation. Both adaptations give significant improvements to the KWS results. By integrating all the techniques, a sports type determination accuracy rate of 92.2% is achieved on the test set consisting of 154 fragments from 17 game programs of ten kinds of sports.