Improved SURE Denoising of Dolphin Whistles with Wavelet Subband Noise Estimation

Denoising dolphin whistles lays the groundwork for subsequent feature extraction and for studying how dolphins adaptively modify their whistles in response to noise interference. To address the challenges of inaccurate noise estimation and residual noise in wavelet thresholding under non-stationary underwater noise, this study proposes an improved Stein’s Unbiased Risk Estimator (SURE) thresholding method. The proposed method innovatively combines impulse noise suppression with a wavelet subband noise estimation strategy based on the integral area of the zero-crossing rate (ZCR). Results demonstrate that the improved SURE method outperforms other threshold denoising techniques. Furthermore, within the same SURE framework, the subband-based noise estimation in the improved method outperforms approaches that rely on fixed noise intervals for threshold selection, providing a more adaptive and accurate preprocessing strategy for dolphin whistle analysis and future research.