| Speaker recognition has become one of the mainstream technologies in the field of Biometric Authentication and been widely applied due to its unique advantages. However, with the appearance of voice change softwares in mobile phones, the electronic disguised voices produced by them has become a big challenge to the current speaker recognition technologies because they cover the personal characteristics in speech. Once used by criminals, the consequences will be severe. Therefore, speaker recognition under the electronic disguised voice may be a hot spot in the field of forensic identification. While, there are three big problems when recognizing the identity of the speakers from the electronic disguised voices, a) how to confirm the disguised degree, b) the system for electronic disguised voice, c) how to get the original speech from the electronic disguised voices. According to the three problems, the main work of the paper is as follows.(1) Proposing a model of DTW combined with VQ which can recognize the identity of the speakers. Choosing the MFCC as the characteristic parameters of electronic disguised voice whose disguised degree is confirmed via dynamic programming. Then, adjust the disguised degree of testing and training speech to a same degree and identify the target speaker.(2) Proposing a method which can distinguish the gender between speakers in the electronic disguised voice when the speech loses gender characteristics under high disguised degree.The disguised degree produced by DTW model will be obviously different under the two situation, First, the gender between testing voice and training voice is same. Second, the gender between testing voice and training voice is different. Using this property, the method to distinguish the gender between speakers is proposed.(3) Realizing a method which can recover the original speech from the electronic disguised voices according to the speech signal resampling and the PSOLA by adjusting the pitch and voice speed dynamicly. |
PDF Full Text Request