Scholarlink

<p>Auditory steady state response (ASSR) signal is an important biometric for performing the authentication process. By reducing the number of electrodes for collecting the ASSR signal, development of authentication systems with real-life applications becomes feasible. However, the signal-to-noise (SNR) ratio of the ASSR data is also negatively impacted, which leads to deteriorating the authentication performance. In order to address this, in this paper, we design a novel self-supervised learning-based scheme using transformers for the task of ASSR signal SNR enhancement. In the development of the proposed scheme, we design a novel optimization process by utilizing regularization terms from the prior information of the ASSR data. The results of various experimentations demonstrate the effectiveness of the proposed scheme in designing high-performance biometric authentication systems. Specifically, the proposed scheme achieves 1.97 dB superior SNR enhancement comparing to the baseline deep learning-based denoising counterpart.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

SNRENN: A Transformer-Based Neural Network with Self-Supervised Learning for Auditory Steady State Response Signal SNR Enhancement

  • 期刊论文

  • 发表日期：

    2025年10月5日

查看全文
  • Alireza Esmaeilzehi,
  • Ensieh Khazaei,
  • Dimitrios Hatzinakos

摘要

Auditory steady state response (ASSR) signal is an important biometric for performing the authentication process. By reducing the number of electrodes for collecting the ASSR signal, development of authentication systems with real-life applications becomes feasible. However, the signal-to-noise (SNR) ratio of the ASSR data is also negatively impacted, which leads to deteriorating the authentication performance. In order to address this, in this paper, we design a novel self-supervised learning-based scheme using transformers for the task of ASSR signal SNR enhancement. In the development of the proposed scheme, we design a novel optimization process by utilizing regularization terms from the prior information of the ASSR data. The results of various experimentations demonstrate the effectiveness of the proposed scheme in designing high-performance biometric authentication systems. Specifically, the proposed scheme achieves 1.97 dB superior SNR enhancement comparing to the baseline deep learning-based denoising counterpart.