Scholarlink

<p>This study proposes a transfer learning-based noise suppression method employing a stacked bidirectional long short-term memory (BiLSTM) network to isolate Southern Resident killer whale vocalizations in complex underwater acoustic environments. The proposed framework employs a two-stage training strategy, first using simplified synthetic waveforms for pretraining and then fine-tuning on field data. Input features are extracted via the fractional Fourier transform (FrFT), enabling improved feature separability. The optimal FrFT angle range was found to be − 67.5° to 67.5°, achieving the best denoising performance, with a peak signal-to-noise ratio (PSNR) of 42.81 dB and a root mean square error (RMSE) of 0.72e–2. Comparative analyses on both synthetic and field datasets confirm that the proposed method effectively suppresses ambient and background noise while minimizing the distortion of essential bioacoustics signals of whales. This transfer learning approach demonstrates superior performance over conventional filtering techniques, which often degrade bioacoustic signals. These findings highlight the robustness of the method and its significant potential for enhancing marine mammal tracking and acoustic monitoring performance in marine science.</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

Effective noise suppression for Southern resident killer Whale recordings using Transfer-Learned stacked BiLSTM

  • 期刊论文

  • 发表日期：

    2026年1月8日

查看全文
  • Seoje Jeong,
  • Wookeen Chung,
  • Dawoon Lee,
  • Jungyong Park,
  • Ho Seuk Bae

摘要

This study proposes a transfer learning-based noise suppression method employing a stacked bidirectional long short-term memory (BiLSTM) network to isolate Southern Resident killer whale vocalizations in complex underwater acoustic environments. The proposed framework employs a two-stage training strategy, first using simplified synthetic waveforms for pretraining and then fine-tuning on field data. Input features are extracted via the fractional Fourier transform (FrFT), enabling improved feature separability. The optimal FrFT angle range was found to be − 67.5° to 67.5°, achieving the best denoising performance, with a peak signal-to-noise ratio (PSNR) of 42.81 dB and a root mean square error (RMSE) of 0.72e–2. Comparative analyses on both synthetic and field datasets confirm that the proposed method effectively suppresses ambient and background noise while minimizing the distortion of essential bioacoustics signals of whales. This transfer learning approach demonstrates superior performance over conventional filtering techniques, which often degrade bioacoustic signals. These findings highlight the robustness of the method and its significant potential for enhancing marine mammal tracking and acoustic monitoring performance in marine science.