Voiceprint Recognition Technology And Its Application In Forensic Identification

Voiceprint recognition (VR) technology, one of the biometric certification technologies, makes use of the speech coefficients in the waveform which represent the speaker's physiological and physical features to identify the speaker automatically. The paper mainly studies the application of voiceprint recognition in the forensic identification, including the status and problems of the forensic identification in China and the development for the VR system according to the demand of the forensic practice.At present, there are three main reasons why more and more voice cases couldn't be identified. First, the bodies of forensic identification only depend costly on foreign advanced equipments to analyze and recognize voiceprint. Second, forensic voice identification is lacking of industry standardization in China. Finally, the uniqueness of the Chinese, the complex of pronunciation, dialect diversity, as well as the effect of tone, bring more difficulties and challenges to VR technology. If only utilizing foreign VR technology is often unable to meet the requirements of Chinese speech recognition. The paper aims to develop a low-cost system which also could meet the requirements of the forensic identification.To meet the demands of forensic practice and adapt to the situation of shortage of funds in medium-sized forensic institutions, Based on the theoretical foundation of VR technology and the status of forensic voiceprint identification, the paper does some researches as follows:(1) Study the existing problems and reasons why more and more voice cases couldn't be identified. Conduct an in-deep research to procedures and standardization of forensic voice identification according to the demands of VR technology.(2) As the complexity of the original speech signal in the judicial case, the pretreatment analysis is very important. The key studies included in this paper, are the distortion compensation, de-noising based-on the wavelet theory, pitch endpoint detection and filtering interference (3) Extract static and dynamic features which better reflect what speaker physiological and behavioral parameters by studying and comparing the existing mainstream voice signal features, such as Mel frequency cepstrum coefficient (MFCC) and its difference (AMFCC).(4) Study the pattern matching algorithm and system design. Mainly study the voiceprint verification system based on the Gaussian mixture model (GMM). The paper completed the feature extraction, GMM training and recognition by MATLAB6.5 and Praat software which also verified the theory of VR and the analysis methods. Besides, it also did some further analysis on the actual case of voiceprint recognition in forensic practice.