The Method And Implementation Of Conformity Assessment Of Audio And Video Information Based On Specific Pronunciation Unit

The problem of aging population is becoming more and more serious in China, the socialsecurity face serious issues of fraud such as impersonator. The authentication problem of theright beneficiary is prominent increasingly; Lip-synching problems always reported in largeconcert, but we can’t get any evidence, so it is necessary to test suspected lip-synching;Animation industry is the low carbon industry, which the state encourage, but quality ofcartoon voice is lack of objective evaluation technology. Because the real voice is producedby the human pronunciation organs, speech signal and lip motion information are of strictconsistency. This paper explore the authenticity of identity authentication and voice speechsamples based on consistency analysis with the audio and video, and improve the accuracy ofidentity authentication of social beneficiaries, and prevent impersonator effectively, as well asproviding technical basis of the objective evaluation of voice quality and lip-synching forsolving the problems.This paper present consistency analysis methods based on specific pronunciation. Thebasic method for the analysis is CoIA, which means the analysis of inertia. Associate themovement of voice and lip motion of video, and do the consistency analysis of voice and lipdynamic in video. It is divided into training phase and test and analysis phase, the trainingstage do the feature extraction of voice and image of lip moving in the video respectively, andfigure out the mapping matrix; The test and analysis stage project the feature onto themapping matrix. The mean value of the covariance of the projection is the correlationcoefficient, The larger of the value of the correlation coefficient of CoIA, the more relevant ofthe audio and video. If the equal error rate is smaller, the consistent evaluation performance ofaudio and video is better. Experiment results show that CoIA method can realize audio andconsistency analysis accurately relatively.The selection of specific pronunciation unit and consistency analysis of video and audiobased on specific pronunciation unit are the innovation points in this paper, selecting specificpronunciation unit for consistency analysis of audio and video alternative sentences. Thispaper analysis the mouth shape of initials and finals in Chinese voice mouth firstly, clusteringthe finals on the basis of the mouth similar characteristics, and cluster a class with the sameshape parameters of the finals. Clustering finals to be a total of16classes eventually.Secondly selecting coefficient pronunciation unit category of high correlation as a particularunit by CoIA method, and conducting consistent data and inconsistent data to do consistencyanalysis through the experiments to validate the rationality of specific pronunciation unit selectd. Finally, do the audio and video consistency analysis with the whole sentences andspecific pronunciation unit extracted from the whole sentences. It is the contrastive analysis ofthe whole sentence and the clustering based on a particular syllable. Experiment databaseinclude350sentences, and the length of one of the whole sentence is between about3seconds to10seconds. Do the extraction and identification of specific pronunciation unitcluster to7groups from a whole sentence by energy and zero crossing rate and thefundamental frequency. The length of a specific pronunciation unit is between about0.3seconds to0.8seconds. CoIA Specific pronunciation unit samples extracted from350sentences. So the length of selection of specific pronunciation unit decrease three-quartersthan of the sentence, so using specific pronunciation unit decrease operation data. Do theaudio and video consistency analysis for specific pronunciation unit and the whole sentenceby CoIA algorithmn respectively. The experimental results show that the error rate ofconsistency evaluation of specific pronunciation unit is2.7%lower than that of the wholesentence.