Ricard Marxer
@ricard.bsky.social
ml, audio, cv, nlp, speech, bioacoustics // Assoc. Prof. at Université de Toulon, researcher at LIS CNRS UMR 7020, director of http://www.master-mir.eu in marine robotics and AI
@brbbfreitas.bsky.social @frantsjensen.bsky.social @arikkershenbaum.bsky.social @malavika-m.bsky.social @hollyhowls.bsky.social
June 20, 2025 at 1:18 PM
@brbbfreitas.bsky.social @frantsjensen.bsky.social @arikkershenbaum.bsky.social @malavika-m.bsky.social @hollyhowls.bsky.social
Robert C. Liu, Malavika Madhavan, Andrew Markham, Marie A. Roch, Holly Root-Gutteridge, Martin Šálek, Grace Smith-Vidaurre, Ariana Strandburg-Peshkin, Megan R. Warren, Matthew Wijers
June 20, 2025 at 12:14 PM
Robert C. Liu, Malavika Madhavan, Andrew Markham, Marie A. Roch, Holly Root-Gutteridge, Martin Šálek, Grace Smith-Vidaurre, Ariana Strandburg-Peshkin, Megan R. Warren, Matthew Wijers
🔗 Project page: www.ricardmarxer.com/f0-bioacoust...
📄 Paper: doi.org/10.1080/0952...
Fantastic work with Paul Best and Marcelo Araya-Salas, Axel G. Ekström, Bárbara Freitas, Frants H. Jensen, Arik Kershenbaum, Adriano R. Lameira, Kenna D. S. Lehmann, Pavel Linhart
📄 Paper: doi.org/10.1080/0952...
Fantastic work with Paul Best and Marcelo Araya-Salas, Axel G. Ekström, Bárbara Freitas, Frants H. Jensen, Arik Kershenbaum, Adriano R. Lameira, Kenna D. S. Lehmann, Pavel Linhart
Benchmarking Neural F0 Estimation for Bioacoustics
Bioacoustic fundamental frequency estimation: a cross-species dataset and deep learning baseline
www.ricardmarxer.com
June 20, 2025 at 12:14 PM
🔗 Project page: www.ricardmarxer.com/f0-bioacoust...
📄 Paper: doi.org/10.1080/0952...
Fantastic work with Paul Best and Marcelo Araya-Salas, Axel G. Ekström, Bárbara Freitas, Frants H. Jensen, Arik Kershenbaum, Adriano R. Lameira, Kenna D. S. Lehmann, Pavel Linhart
📄 Paper: doi.org/10.1080/0952...
Fantastic work with Paul Best and Marcelo Araya-Salas, Axel G. Ekström, Bárbara Freitas, Frants H. Jensen, Arik Kershenbaum, Adriano R. Lameira, Kenna D. S. Lehmann, Pavel Linhart
We provide:
🔧 A tool with pretrained models to estimate F0 on new data
📊 An annotated dataset of 250k+ vocalisations
🔧 A tool with pretrained models to estimate F0 on new data
📊 An annotated dataset of 250k+ vocalisations
June 20, 2025 at 12:14 PM
We provide:
🔧 A tool with pretrained models to estimate F0 on new data
📊 An annotated dataset of 250k+ vocalisations
🔧 A tool with pretrained models to estimate F0 on new data
📊 An annotated dataset of 250k+ vocalisations
Key findings:
✅ Neural network models show strong potential
✅ Supervised models generalise to unseen species
✅ Self-supervised models can perform on par with supervised ones in some cases
✅ As expected, performance correlates with F0 saliency
✅ Neural network models show strong potential
✅ Supervised models generalise to unseen species
✅ Self-supervised models can perform on par with supervised ones in some cases
✅ As expected, performance correlates with F0 saliency
June 20, 2025 at 12:14 PM
Key findings:
✅ Neural network models show strong potential
✅ Supervised models generalise to unseen species
✅ Self-supervised models can perform on par with supervised ones in some cases
✅ As expected, performance correlates with F0 saliency
✅ Neural network models show strong potential
✅ Supervised models generalise to unseen species
✅ Self-supervised models can perform on par with supervised ones in some cases
✅ As expected, performance correlates with F0 saliency
The diversity of bioacoustic signals—in frequency range, harmonicity, SNR, etc.—makes F0 estimation particularly challenging.
We study how well #MIR F0 estimation transfers to non-human vocalisations, benchmarking SOTA models across 14 taxa:
We study how well #MIR F0 estimation transfers to non-human vocalisations, benchmarking SOTA models across 14 taxa:
June 20, 2025 at 12:14 PM
The diversity of bioacoustic signals—in frequency range, harmonicity, SNR, etc.—makes F0 estimation particularly challenging.
We study how well #MIR F0 estimation transfers to non-human vocalisations, benchmarking SOTA models across 14 taxa:
We study how well #MIR F0 estimation transfers to non-human vocalisations, benchmarking SOTA models across 14 taxa: