Study On Characteristic Parameters Of The Deliberated Imitation Speech

People's voice is independent, but still can be imitated. The deliberate imitated voice by the imitator can not be recognized by the listeners when the similarity is high and even some identity verification systems will be deceived, which does harm to information security. Therefore, it is necessary to analyze the characteristics of the deliberated imitated voice and find out its characteristic parameters. This will have a broad application prospect in the criminal investigation and national defense.Firstly, this paper introduces the research status of imitation speech recognition and the concept of deliberate imitation. The factors that affect the imitated speech's similarity are analyzed, and the information security caused by deliberated imitation speech is also pointed out. Then the commonly used characteristic parameters in speaker recognition are introduced such as pitch, LPCC and MFCC, as well as their dynamic parameters. Based on the research of characteristic parameters, speaker recognition method based on VQ is introduced and a new speaker recognition method based on the novel spectral distance and maximum a posteriori vector quantization(VQ-MAP) procedure are proposed which is verified that it is fast and efficient according to the results of experiment on speaker recognition system.Secondly, the paired-comparison method is used for subjective evaluation of the similarity of the deliberated imitation speech and difference of characteristic parameters is computed by Euclidean distance and novel spectral distance. The deliberated imitation speech database is established for experiment and the paired-comparison method is used to get the sort of similarity of subjective evaluation, then to calculate the difference ratio of characteristic parameters. The principle distinguished the original speech and deliberated imitation speech is that the inter-speaker distance between the original speech and deliberated imitation speech is larger than the intra-speaker distance of the original speech and the deliberated imitation speech and this principle is utilized in individual performance of characteristic parameters. Then the sorts of subjective evaluation and the difference of characteristic parameters are compared to see whether the sorts of them are same. Based on this, the distance ratios of different characteristic parameters are compared and we conclude characteristic parameters that can be used for recognition of the deliberated imitation speech and the performance of characteristic parameters.Finally, we sum up this research and propose future research directions.