Design And Implementation Of Judicial Speech Identity Identification System

With the advancement of science and the continuous development of network technology,the number of crimes that using computer and network as crime tools has increased significantly,and the electronic data generated by them has become one of the hot issues in judicial identification research.As one of the most convenient judicial evidences in electronic data,voice electronic data occupies a considerable proportion in judicial evidence,which makes the theoretical significance and practical value of automated speech recognition technology getting more and more attention.Based on the design and implementation of the judicial speech identity authentication system,this paper deeply studies the relevant theories and techniques of speech recognition by consulting a large number of relevant domestic and foreign literatures and designs a set of intelligent speech identity identification system which is suitable for judicial authentication.The system is mainly divided into two parts: speakers segmentation clustering and speech identity identification.The former is to preprocess the sample speech,and the latter is to identify the speaker identity of the target speech after processing.The main work contents of the system are as follows:Firstly,in view of the characteristics of judicial speech materials,this paper uses the famous K-means clustering algorithm to realize speaker segmentation clustering.Then,based on this,an improved spherical K-means(SPK)clustering algorithm is proposed to improve the clustering effect.Finally,Experiments show that the improved spherical K-means(SPK)clustering algorithm has better classification and anti-interference than the general K-means clustering.Secondly,In view of the fact that MFCC features are generally selected for speaker feature extraction in speech identity identification,this paper proposes a new feature extraction method based on the source filter model.According to the speech filter generation model,the LPC coefficients expressing channel features are first extracted,and then the excitation signals in the source filter model are filtered by subband sensing according to the division of Bark subbands.Finally,the combination of LPC coefficients and subband filter energy features is used as a new speaker recognition feature,which solves the problem that traditional MFCC features are susceptible to noise in feature extraction.Then the new feature is introduced into the more widely applicable GMM-UBM model for speech identity identification.The experiment proves that the error rate(EER)of the system is reduced by 30.69% compared with the traditional MFCC feature.Thirdly,combining the proposed improved SPK clustering algorithm and the source filter model feature extraction method,this paper constructs a judicial speech identity authentication system based on the GMM-UBM model.The visualization module of speech signal is added to the system,and by detecting the resonance peak,the user can observe the speech characteristics of different speakers intuitively,so as to play the role of artificial auxiliary identification to ensure the accuracy of the identification.