The Voiceprint Recognition Method Based on the ISCL Network

Voiceprint recognition is a method for discerning the identity of a person by extracting inherent characteristics from their vocal sounds, and it finds widespread applications in security authentication and privacy protection. In order to enhance the performance of voiceprint recognition, this paper proposes a voiceprint recognition approach based on the Improved SimSiam Contrastive Learning (ISCL) network. This method initially separates noise features from vocal characteristics in voiceprints using a novel wavelet transform based on a hybrid threshold function, reducing noise interference and generating spectrograms. Subsequently, these spectrograms are fed into the ISCL network, and an improved infoNCE loss function is employed to enhance the contrastive effects of voiceprint features and the performance of the contrastive network, ultimately achieving voiceprint recognition. Experimental results demonstrate that incorporating wavelet transform and the infoNCE loss function into the contrastive network significantly improves the accuracy of voiceprint recognition. The use of the hybrid threshold function reduces the adverse effects of noise on the network.