Vincent Sirius Kather | Computational Bioacoustics

About Me

I develop machine learning frameworks to analyze complex biological soundscapes. My research focuses on investigating deep learning models beyond their training capabilities, specifically aiming to use models to tackle large unlabeled long-term passive acoustic monitoring (PAM) datasets.

Research Interests

Deep learning: Feature extraction and classification using deep learning models.
Evaluation of high-dimensional spaces: Evaluating embedding spaces using clustering and linear probing.
Ecoacoustics: Assessing how biodiversity can be quantified using automatic processing of audio signals.
Connecting bioacoustics and computer science: Making state-of-the-art deep learning models accessible for ecologists and conservationists through interactive tools.

Publications

Clustering and Novel Class Recognition: Evaluating Bioacoustic Deep Learning Feature Extractors. Vincent Kather, Burooj Ghani, Dan Stowell. (2025). Preprint/Other. [DOI]
Development of a machine learning detector for North Atlantic humpback whale song. Vincent Kather, Fabian Seipel, Benoît Bergès, G. E. Davis, Catherine A. Gibson, M J Harvey, Lea‐Anne Henry, Andrew Stevenson, Denise Risch. (2024). The Journal of the Acoustical Society of America. [DOI]
Interaction of equivalence ratio fluctuations and flow fluctuations in acoustically forced swirl flames. Vincent Kather, Finn Lückoff, Christian Oliver Paschereit, Kilian Oberleithner. (2021). International Journal of Spray and Combustion Dynamics. [DOI]

Software & Open Source

I developed a repository and PyPi package called bacpipe (bioacoustic collection pipeline). The aim is to streamline the usage and evaluation of bioacoustic deep learning models to make sure the state-of-the-art in model development also reaches the bioacoustic experts who benefit most from using them.

During my masters thesis I worked on developing a machine learning model for North Atlantic humpback whale song. I created a repository which includes a graphical user interface to reduce the coding knowledge necessary to run the model: acodet.