Broadcast Advertisement Detection Based On Audio Content Repeatability

Nowadays,broadcasting is an important way for people to receive news,songs,entertainment programs and other information.Because broadcasting has a lot of audience and great influence,it has important practi cal significance to detect advertisements in broadcast.Detection of advertisements is helpful to government supervision,media operations,and advertising.this paper proposes a method of advertisement detection in broadcast based on self-similarity matrix.The main idea of the method is to determine the advertisement segment by calculating the similarity of the audio content in the broadcast in combination with the characteristic that the same advertisement in the broadcast has a large number of frequencies in a fixed period of time.The method proposed in this paper has the characteristics of fast calculation speed and no need to build an advertisement database.The main work of this article is as follows:1.Calculate the self-similarity matrix of the target audio.The audio segments with similar content repetition in the audio are obtained as similar segments through the self-similarity matrix.The final advertisement and starting and ending time are obtained by eliminating the similar segments that appear several times.2.Use an audio classification and segmentation method to optimize the advertisement detection algorithm.Using support vector machines to classify unknown audio,it is divided into four categories:voice,noisy,speech,background sound and music.The classification results obtained by classification of suppor t vector machines are smoothed,and audio segmentation points are obtained based on the smoothed results.3.Use gradient boosting tree to classify advertisement aud io and non-advertisement audio.Audio detection algorithms based on audio repetitiveness have limitations and non-recurring advertisement can not be detected by this method.Therefore,this article uses a advertisement detection algorithm based on gradient boosting tree to detect non-recurring advertisement in audio segments.Finally,combine the detection results of the two methods.