Voice spoofing detection method based on deep residual shrinkage network

Abstract

The invention discloses a voice spoofing detection method based on a deep residual shrinkage network, and the method comprises the steps: firstly carrying out the preprocessing of a to-be-detected voice, carrying out the transformation of the preprocessed voice feature data, and obtaining a corresponding constant Q cepstrum coefficient feature, a Mel-frequency cepstrum coefficient feature and a spectrogram feature; respectively processing the constant Q cepstrum coefficient feature, the Mel frequency cepstrum coefficient feature and the spectrogram feature by adopting a deep residual shrinkage network to obtain three corresponding depth features; respectively inputting the three depth features into a deep neural network classifier, and calculating to obtain detection scores corresponding to the three depth features; finally, the detection scores corresponding to the three depth features are fused, and whether the voice to be detected is real voice or not is judged. According to the method, the discriminant feature learning ability in a complex acoustic environment is improved, the generalization of the system is improved, and the application scene is wider.

Description

technical field [0001] The application belongs to the technical field of voice detection and deep learning, and in particular relates to a voice deception detection method based on a deep residual shrinkage network. Background technique [0002] In recent years, the role of biometric-based identity authentication technology in data security and passability authentication has become more and more important. Due to the development of acquisition and sensing equipment, automatic speaker verification technology has received extensive attention and has been applied to smart device login, access control, online banking, etc. However, various voice forgery technologies threaten the security performance of automatic speaker verification systems. Four types of forged voice spoofing attacks have been identified: speech synthesis, voice conversion, voice imitation, and replay, which can generate fake voices similar to legitimate user voices. voice. Logical access attacks, which focus...

AI Technical Summary

Problems solved by technology

[0004] Although the training performance of existing methods has been improved, they encounter unknown types of attacks in practical applications, and these attacks usually have a different statistical distribution from known attacks, resulting in a large performance gap between training and application, which shows that The generalization ability of deception detection system to unknown attacks still needs to be improved

In addition, due to the presence of noise, reverberation, and channel interference in real environments, various types of deception detection systems suffer from significant performance degradation in the face of complex acoustic environments

Images