Research On Pornographic Audio Detection Algorithm Using MFCC Features And Vector Quantization

With the super-fast development of internet, the spread of unhealthy culture on the internet more and more increasingly drew government and people’s attention. Even though the Chinese government already put a large investment (many people and large fund) to resist the unhealthy culture to spread, because of the virtue of internet and the huge of resource of internet, there are many difficulties to detect and prevent unhealthy media spreading. Eroticism, violence and other unhealthy network culture are spreading with text, picture, video, even some unhealthy information is spreading by computer games. And the methods of spread are varied. The audio signal is one of the methods of unhealthy spread. Research on the pornographic audio detection is always important research area for many scientists. With the information quickly develop and deeply research, audio detection and audio recognition are more and more applied for medical device. This paper presents an algorithm for detecting whether an audio which is extracted from the video by FFMPEG is health or unhealthy, at the same time as an auxiliary means of video detection, it improves the efficiency of video detection.This paper is focus of the algorithm of detection to provide a new algorithm of detection based on the Mel Frequency Cepstral Coefficents (MFCC). As an audio of the same classification, the characteristic parameters are similar, there are three important steps of this algorithm:first of all, it will provide a method to extract the audio signal from the video stream to be pre-processed. And then extract the features from the processed audio signal. At the last the features vector will be trained an initial vector code, set a threshold to incessantly iterate vector code until get a wonderful vector code for the unhealthy audio signal. Compare the testing sample with the vector and gain a score of the audio to degust the nature of the audio. This algorithm provides accuracy for audio detection. This algorithm is implemented in the matlab and VC++6.0.