Echo lite voice fusion network: advancing underwater acoustic voiceprint recognition with lightweight neural architectures

Jiaqi Wu, Donghai Guan, Weiwei Yuan

Abstract: Underwater acoustic voiceprint recognition, serving as a key technology in the field of biometric identification, presents a wide range of application prospects, especially in areas such as marine resource development, underwater communication, and underwater safety monitoring. Conventional acoustic voiceprint recognition methods exhibit limitations in underwater environments, prompting the need for a lightweight neural network approach to optimally address underwater acoustic voiceprint recognition tasks. This paper introduces a novel lightweight voicing recognition model, the Echo Lite Voice Fusion Network (ELVFN), which incorporates depthwise separable convolution and self-attention mechanism, and significantly improves voicing recognition performance by optimizing acoustic feature extraction technology and hierarchical feature fusion strategy. Concurrently, the computational complexity and parameter quantity of the model are substantially reduced. Comparative analyses with existing acoustic voiceprint recognition models corroborate the superior performance of our model across multiple underwater acoustic datasets. Experimental results demonstrate that ELVFN outperforms in various evaluation metrics, notably in terms of processing efficiency and recognition accuracy. Finally, we discuss the application potential and future development directions of the model, providing an efficient solution for underwater acoustic voiceprint recognition in resource-constrained environments.