원문정보
초록
영어
This paper focuses on audio event detection in tennis match. A new audio event detection method is proposed my making use of contextual information. This method detects the sounds of ball hit and line judge’s shout by the grunts of players when they hit the ball. To model for audio events, the proposed method adopts unsupervised learning to use the information of grunts. Compared with current methods, the proposed method possesses two advantages. First, it does not need any labeled data which is appeal in real applications. Second, this method can reduce the mismatch between training and testing to further improve the performance. Experimental results show that the proposed method can improve the performance of audio event detection substantially.
목차
1. Introduction
2. Audio Event Detection Method
2.1. The Positioning of Player’s Shouts
2.2. Method to Find the Candidate Striking Position
2.3. Event Modeling and Confidence Coefficient Selecting
2.4. Detection of the Shouts Caused by umpires’ Ruling
3. Experiment and Analysis
3.1. Experiment Data
3.2. Experiment Setting
3.2. Experiment Results and Analysis
4. Conclusion
Acknowledgement
References