Multimedia Files Repetitive Detection Based On Audio

With the arrival of "cloud storage" and "big data" era,the amount of network data is growing with an incredible speed,among them,audio and video data occupy a high percentage.Some of these network audio and video data are repeated,and these repeated data has brought about many problems in management to these audio and video website.To solve these problems just rely on human is almost impossible to finish,therefore,the study of multimedia files repetitive detection is very important.At present the multimedia files repetitive detection research mainly divided into based on text,based on image and repetitive detection based on audio.The method based on text description is too subjective and for malicious deleted files it cannot be helped.The method based on image can't be work on pure audio files,and its low detected speed and low accuracy cannot satisfy the reality request.Therefore,we mainly studied the multimedia files repetitive detection method based on the audio.This paper mainly completed the following work:Firstly,this paper studies the feature of the audio.We introduced the time domain features the frequency domain features and the cepstrum domain features of the audio.We mainly studied the audio feature of mel frequency cepstrum coefficient(MFCC),pitch class profiles(PCP),mel pitch class profiles(MPCP)and MP who developed on the basis of MFCC and PCP.Secondly,this paper studies the process of model-free repetitive detection based on audio,including multimedia file format unification processing,feature representation,sliding window,cross recurrence plot and repetitive decision.Format unification process is to extract audio data from multimedia files and decode it.Feature representation refers to using one of four audio features respectively to replace the audio data.Sliding window is used to solve the problem when multi media files' size is different.Cross recurrence plot refers to analyzing two pieces of audio features' similarity based on the 2-dimensional figure.Repetitive decision refers to judging if two multimedia files are repeated or not by compared the similarity with the threshold.This paper compares the four audio features(MFCC,PCP,MPCP,MP)experimental performance and through the experiment we found that model-free repetitive detection system based on audio has high accuracy.Finally,this paper studies the process of model-based repetitive detection based on audio,including multimedia file format unification processing,feature representation,sliding window,cross recurrence plot,feature vector extraction based on the cross recurrence plot,Max Pooling and support vector machine(SVM).This system regarded the repetitive detection as binary classification problems,and then using the SVM model.Through the experiment we found that the model-based audio repetitive detection based on SVM model has a better performance.