Research On Audio Advertising Detection Algorithms

Rapid detection and accurate localization of target advertising from a lot of TV programs can make great sense for media operation department, advertising companies and even media regulator. With the progress of multimedia technology and the advancement of commercialization, increasingly advanced multimedia information detection technology, including image detection, video and audio detection techniques, has been widely used in the detection and monitor work of TV advertising. In this paper, audio detection technology is used for the detection and localization of target advertising on the basis of audio signal analysis of television advertising.Because all the audio ads on TV are carried by kinds of media, we need to do preprocessing to get unified format audio files. Moreover, the time of TV advertising has certain limitation, the length of sample audio advertising used for detection is relavtively fixed. Due to the specific detection goal, it belongs to the fixed audio test category. On the basis of consulting a large number of literature and quantities of experiments, this paper comes up with audio advertising detection algorithm from whole form of audio signal waveform, local content level characteristics and audio information by three different aspects. The algorithm both pays attention to detection efficiency and detection performance. And some audio television advertising data provided by a TV station has been used for the verification of the algorithm performance. The main content of the paper is as follows:1. Coming up with audio advertising detection algorithm based on signal similarity for the purpose of audio advertising detection. According to the signal similarity theory in digital signal processing, the algorithm only need to calculate the similarity coefficient between equal length sample audio signal and audio signal to be detected for signal match and detection. In order to improve the rapidity and adaptability, the algorithm is improved by convolution theorem in circle convolution theory so as to achieve fast and efficient detection of audio advertising.2. Using the characteristics in the time domain, frequency domain and the coefficient domain which is extracted form the audio signal, and based on short-time stable characteristics of audio signal, an audio advertising detection algorithm based on audio frame feature matching is put forward. Due to the low decetion efficiency of the direct detection algorithm, as well as the commonly used histogram pruning algorithm is easy to lose the sequence information of audio frame and has poor precision, this paper presents a layered detection algorithm based on histogram improvement. In this algorithm, the testing process is divided into two stages, namely rough detection and precision detection. In the first stage, it uses histogram algorithm for screening of candidate audio clips, while the precision stage uses the strict matching method to detect the target fragment which is the same as the sample audio, and then the results are got. Experiment proves that the layed detection algorithm based on histogram improvement not only ensure the good detection effect, but also can shorten the detection time.3. Compressive sensing(CS) theory as a relatively new theory in the field of signal processing is introduced, and by utilizing the audio signals sparsity on the specific transform domain, an audio advertising detection algorithm based on compressive sensing theory is proposed in this paper. Compressive sensing technology can achive dimension reduction of the audio signal by compression observation, and makes full use of the key information in the signals, so as to realize the matching of the detection in the information level. According to the experiment results, the compressive sensing theory and technology can be applied to the audio detection filed. Compared with the traditional detection algorithm based on feature matching, this detection algorithm is more efficient and has better detection performance.