Deep learning methods for wildlife bioacoustics and ecology
IJCNN 2020 Special Session
Organisers: Dan Stowell, Veronica Morfi, Ricard Marxer
Extended deadline: 30th January
We are pleased to invite submissions for a Special session on "Deep learning methods for wildlife bioacoustics and ecology" at the International Joint Conference on Neural Networks (IJCNN 2020), running as part of the IEEE World Congress on Computational Intelligence (WCCI 2020).
The increasing impact of human activity on the Earth's ecosystems has led to the massive loss and fragmentation of natural habitats. Together with human effects on the climate system, this change has accelerated the extinction of several species and caused the endangerment of many ecological processes.
The complexity and increasing fragility of the interactions between humans and nature require making use of new sources of evidence, if we are to face emerging environmental challenges and take into account legacy effects. One recent approach to deal with these challenges is ecoacoustics. This area of research focuses on studying the soundscape—the acoustic footprint of an environment including its plants and animals—and is a source of a vast amount of information that could be used efficiently in monitoring schemes. The application of soundscape analysis could enable us to efficiently investigate the dynamics of animal behaviour, particularly when habitats are modified, fragmented, or destroyed.
Soundscape analysis has a very different focus than most speech technology, since it is crucial to deal with highly varying sound environments having multiple sound sources of different types, often distant and overlapping.
In the last decade, deep learning approaches have radically changed several research fields. In bioacoustics and ecoacoustics, deep learning methods have achieved unprecedented results, outperforming previously used traditional signal processing methods for e.g. species classification. Partial progress has also been made in assembling large ecoacoustic datasets. However, despite some success, there still exist several open challenges. Natural ecosystems often prove difficult to study by current methods, due to the difficulty of collecting data, the complexity of habitats and highly varying noise. All these make automatic monitoring of soundscapes a challenging task.
Hence it is important for the scientific community to better understand the extent to which deep learning approaches can be efficiently employed in automatic soundscape monitoring, and to use these application scenarios to motivate new methodological developments. The aim of this session is to bring together people that can discuss and present the most recent advances in this field.
The organisation of this session builds on the organisers’ previous experience hosting sessions on related topics at zoological conferences (Int Bioacoustics Congress 2017, 2019) and signal processing conferences (ICASSP 2019; EUSIPCO 2017, INTERSPEECH 2016), which demonstrated the research interest and the contributions of many research groups. This session aims to bring the focus specifically to the deep learning community, and to build solid understanding of how deep learning methods can be better designed and deployed to address these globally important questions about humanity’s natural environment.
- Deep learning for statistical ecology (population estimation)
- Detection/classification of animal sounds
- Individual identification (similar to speaker recognition) for wildlife/animal sound
- Sound source separation in natural sound scenes
- Acoustic localisation
- Deep learning for analysis of natural sound scenes
- Deep learning models of animal perception
- Modelling multi-agent acoustic interaction
- Audiovisual/multimodal natural signals
- Cross-modal deep learning
- Paper Submission Deadline: ~~15 January 2020~~ extended to 30 January 2020
- Paper Acceptance Notification Date: 15 March 2020
- IEEE IJCNN 2020: 19-24 July 2020
Special session manuscripts should be submitted using the IJCNN 2020 online submission system, and will be subject to the same peer-review process. All accepted papers will be included in the regular conference proceedings.
Dan Stowell is a Lecturer in machine listening at Queen Mary University of London. He co-leads the Machine Listening Lab at Queen Mary University of London, based in the Centre for Digital Music, and is also a Turing Fellow at the Alan Turing Institute. Dan has worked on voice, music and environmental soundscapes, and recently led a five-year EPSRC fellowship project researching the automatic analysis of bird sounds. He was lead organiser of VIHAR 2019 (the international workshop on Vocal Interaction in-and-between Humans Animals and Robots), and has been chair of many special sessions on machine learning and audio at leading international conferences: ICASSP 2019 (Int Conf Acoustics Speech and Signal Processing), ICEI 2018 (Int Conf Ecological Informatics), EUSIPCO 2017 (European Signal Processing Conf), IBAC 2017 and 2019 (Int Bioacoustics Conf). http://mcld.co.uk/research/
Veronica Morfi is a postdoctoral research assistant in the Machine Listening Lab at Queen Mary University of London (QMUL). During her PhD she created new deep learning methods for effective analysis of low-resource bioacoustic data, and also completed a research project on speech technology with Samsung R&D UK. In 2018 she was Chair of the IEEE Student Branch for QMUL.
Ricard Marxer is a Maître de conférences (Asst. Prof.) at the Université de Toulon and a researcher at the CNRS Laboratoire d’Informatique et Systèmes. His research interests include speech perception and processing, music analysis, source separation, machine listening and unsupervised learning. He was organiser of an InterSpeech 2016 special session “Intelligibility under the Microscope”, and has been co-organiser of many other events, including two CHiME speech separation and recognition challenges, and VIHAR 2017 and 2019 (the international workshop on Vocal Interaction in-and-between Humans Animals and Robots). http://www.ricardmarxer.com/