Disguised voice detection method based on complete and local binary patterns

Abstract

The invention relates to a disguised voice detection method based on complete and local binary patterns. The disguised voice detection method comprises the steps of carrying out variable Q conversionon all voices in a real voice library and a corresponding disguised voice library, so as to obtain voice spectrograms of all true voices and disguised voices; converting the voice spectrograms into corresponding grayscale images, and generating corresponding texture characteristics by processing the grayscale images through the complete and local binary patterns; obtaining a support vector machinethrough training by using all the texture characteristics as a training set for training the support vector machine; and inputting a to-be-recognized voice to the support vector machine so as to recognize the disguised voice. The voice spectrograms obtained through variable Q conversion are clearer in texture, and beneficial to extracting the texture characteristics of voice signals; moreover, the complete and local binary patterns are adopted, local symbol difference value information and local amplitude difference value information of the voice spectrograms are included, the texture characteristics of the signals can be obtained more comprehensively, the classification by the support vector machine is facilitated, and the accuracy of recognizing the disguised voice is improved.

Description

technical field [0001] The invention belongs to the technical field of speech recognition, and in particular relates to a method for detecting a fake speech based on a complete partial binary pattern. Background technique [0002] Masquerade voice detection is to analyze the speaker's voice, and then identify whether it is the real speaker's voice or the artificially maliciously disguised voice. Masquerading voice is usually generated by device playback, voice conversion and speech synthesis technology. Through these deliberate operations, it can be disguised as a specific speaker's voice, so as to deceive the speaker recognition system. The masquerade speech recognition system can realize anti-masquerade detection for malicious masquerade speech, improve the security performance of the speaker recognition system, and has broad application prospects. Masquerade speech recognition usually needs to extract the features of the target speech signal, and then compare and analyze...

AI Technical Summary

Problems solved by technology

In practice, because the speaker recognition system has to deal with various unknown disguised voice attacks, the detection effect based on the above characteristics is often not very ideal.

Images