Research On ADs Monitoring Technology Based On Audio Match

With the rapid development of Chinese broadcast television and multi-media technology, the television broadcast audio data has increased exponentially. It brought a huge challenge for media assets management and supervisor. In order to meet the demand of the broadcast television content monitoring under the new situation, to establish and perfect the broadcast and TV monitoring system and realize the automatic monitoring is increasingly urgent. Therefore the audio matching technology based on content become a hotspot of research on radio monitoring, which has important research significance and application value.In this thesis, according to the actual demand of radio advertising monitoring, we design and implement a broadcasting ADs monitoring system, based on studying and improving the existing audio retrieval algorithm. The system can quickly and correctly implement the radio ads audio monitoring, statistics and feedback to the customer. The specific content and the innovation work is demonstrated as follows:1. Radio broadcast always needs to use tuner to make EQ processing for polishing. And the broadcast audio is vulnerable to noise interference and time stretching attack leading to distortion. Existing fingerprint extraction algorithm does not balance robustness and accuracy well, so this article adopts MCLT-based improved spectra, through dynamic threshold value to determine the peak points to establish audio fingerprint, to improve audio fingerprint of noise resistance and EQ treatment resistance. Using audio fingerprint, appropriate hash function is used to construct the fingerprint hash to enhance fingerprint hash the time stretching attack resistance. The experimental results show that comparing with the existing methods, the algorithm has a stronger resistance to noise and robustness, especially for time stretching attack.2. Traditional advertising monitoring method is to adopt the method of template matching. The method can accurately locate the advertising audio start-over time. The broadcast audio stream is as long as 24 hours, while usual advertising audio is only about 10 s. In consequence, the traditional template matching method locates the time less efficiently. This thesis puts forward a kind of audio matching method based on statistics, which gets time information from the collision of hash point, rules out the corresponding relationship between query audio and target audio by least square analysis. Experiments show that this method can accurately locate the target audio information quickly, meet the actual demand of monitoring.3. Combined with the improved audio retrieval algorithm and the actual functional requirements, a radio advertising monitoring system was designed and implemented. The system completes audio ADs monitoring using improved audio fingerprint extraction algorithm and the method of audio matching. The system mainly includes an audio comparison analysis subsystem and an advertising business processing subsystem. It can realize radio advertising monitoring full-automatically. At the same time, the system can be easily transplanted to other radio audio monitoring business.