% pubman genre = conference-paper
@inproceedings{item_3286386,
title = {{ORCA-CLEAN: A deep denoising toolkit for killer whale communication}},
author = {Bergler, Christian and Schmitt, Manuel and Maier, Andreas and Smeele, S. and Barth, Volker and Noth, Elmar},
language = {eng},
issn = {1990-9772},
doi = {10.21437/Interspeech.2020-1316},
publisher = {International Speech Communication Association},
year = {2020},
date = {2020-10},
abstract = {{In bioacoustics, passive acoustic monitoring of animals living{\textless}br{\textgreater}in the wild, both on land and underwater, leads to large data{\textless}br{\textgreater}archives characterized by a strong imbalance between recorded{\textless}br{\textgreater}animal sounds and ambient noises. Bioacoustic datasets suffer{\textless}br{\textgreater}extremely from such large noise-variety, caused by a multitude{\textless}br{\textgreater}of external influences and changing environmental conditions{\textless}br{\textgreater}over years. This leads to significant deficiencies/problems concerning the analysis and interpretation of animal vocalizations{\textless}br{\textgreater}by biologists and machine-learning algorithms. To counteract{\textless}br{\textgreater}such huge noise diversity, it is essential to develop a denoising{\textless}br{\textgreater}procedure enabling automated, efficient, and robust data enhancement. However, a fundamental problem is the lack{\textless}br{\textgreater}of clean/denoised ground-truth samples. The current work{\textless}br{\textgreater}is the first presenting a fully-automated deep denoising approach for bioacoustics, not requiring any clean ground-truth,{\textless}br{\textgreater}together with one of the largest data archives recorded on{\textless}br{\textgreater}killer whales (Orcinus Orca) {\textendash} the Orchive. Therefor, an approach, originally developed for image restoration, known as{\textless}br{\textgreater}Noise2Noise (N2N), was transferred to the field of bioacoustics, and extended by using automatic machine-generated binary masks as additional network attention mechanism. Besides{\textless}br{\textgreater}a significant cross-domain signal enhancement, our previous{\textless}br{\textgreater}results regarding supervised orca/noise segmentation and orca{\textless}br{\textgreater}call type identification were outperformed by applying ORCACLEAN as additional data preprocessing/enhancement step}},
booktitle = {{Interspeech 2020}},
pages = {1136--1140},
address = {wolrdwide},
note = {Interspeech 2020},
}