Self-supervised denoising model based on deep audio prior using single noisy marine mammal sound sample.