In today’s world, with the increasing use of biometric speech recognition systems, voice spoofing attacks have become a major security threat. Spoofing attacks, such as speech synthesis and speech conversion, can compromise the security of the biometric speech recognition system by making it accept fake voices as genuine ones. To address this issue, Junxiao Xue and his team proposed a new method of voice spoofing detection based on physiological-physical feature fusion.
The researchers published their study in Frontiers of Computer Science, where they proposed a novel voice spoofing detection technique that outperforms the existing methods. They combined physiological and physical features in their proposed model that increased the efficiency of feature transmission.
The proposed technique involved a feature extractor, a densely connected convolutional neural network with squeeze and excitation blocks, and a feature fusion strategy. First, the physiological features in the audio were extracted from a pre-trained convolutional network. Second, they extracted the physical features using a densely connected model that was highly parametric-efficient. Finally, the two features were integrated into the classification network for voice spoofing detection.
Compared to the existing single methods, the tandem decision cost function and equal error rate scores improved by 5% and 7%, respectively. The proposed model also showed better performance on both EER and t-DCF compared to the existing models. The team further evaluated the performance of baseline models that combined face features and found that different baseline methods demonstrated varying degrees of performance improvement with the face features.
Future work may focus on extracting more accurate face features and studying more effective feature fusion strategies to detect spoofing attacks. The researchers’ proposed method shows promising results in the field of automatic voice spoofing detection.
Overall, this study highlights the importance of developing robust biometric security technologies that can prevent voice spoofing attacks. With the increasing use of biometric speech recognition systems by governments, businesses, and individuals, developing anti-spoofing measures that incorporate both physiological and physical features can enhance the ability of systems to detect and prevent voice spoofing attacks and maintain the security of biometric speech recognition systems.
<< photo by Andrea Piacquadio >>
