Modified Fast advertisement Detection

This paper puts forward a novel two-steps algorithm to search and detect advertisement signal quickly through broadcast audio signal. It is an improved method based on the Fast Commercial Detection method. The original method takes fixed shift step, and does not take random starting positions of advertisement into account, which may influence the similarity. The modified algorithm in this paper solves the problem by changing the length of frame shift. It is divided into two steps. The first step is rough retrieval which is on the basis of the idea of segmenting audio energy envelope into units. It can pick out the obviously matched signal. The second step is fine retrieval to detect some more complicated situation. Experiment results show that the proposed method can be more accurate.