Göksenin Yüksel

Donders Institute for Brain, Cognition and Behaviour

Radboud University

Nijmegen, the Netherlands

Dutch citizen (EU)

I’m a PhD candidate in AI at the Donders Institute, Radboud University, advised by Kiki van der Heijden and Marcel van Gerven, and a member of the ELLIS Unit Nijmegen. I work on machine hearing: teaching machines to understand sound scenes the way people do, both what they hear and where it comes from. My research combines representation learning with (spatial) audio, building self-supervised audio foundation models that hold up in real, noisy and reverberant environments, and extends to spatial audio large language models.

Most audio foundation models are trained on single-channel audio and are effectively “spatially blind.” My work closes this gap. AudioSphere (NeurIPS 2026) is the first self-supervised audio model to perform sound event localization and detection from a frozen backbone, and it works across microphone array geometries without retraining. Bearings learns soundfield embeddings that plug into frozen single-channel audio encoders without retraining, and EquiSELD makes rotation-equivariant localization networks efficient to train. GRAM learns spatial representations in binaural and Ambisonics formats. I also build open benchmarks (NatHEAR, RealSELD) for evaluating audio models in noisy, reverberant scenes.

My work extends to speech representation learning. SpeechJEPA learns speech representations entirely in latent space, without the clustering that earlier speech JEPAs relied on. Trained in 64 GPU-hours, it outperforms wav2vec 2.0, HuBERT and WavLM at every labeled-data budget for speech recognition. WavJEPA applies the same kind of semantic learning to general-purpose audio, straight from raw waveforms. I train these models on Snellius, the Dutch national supercomputer, with three compute grants totaling 3 million SBUs (about 15,000 H100 GPU-hours). Code and models are on the HAMLET lab GitHub, my personal GitHub and Hugging Face.

Before my PhD, I worked at ScreenPoint Medical as a software engineer on large-scale medical AI software. During a research internship there, I developed an MRI segmentation pipeline for breast density estimation, which led to a first-author paper at MIDL 2024. During my bachelor’s, I interned as a data scientist at Arcadis, building machine learning dashboards for housing price analysis and EV charging placement. I hold an MSc in AI from the University of Amsterdam and a BSc in AI from Radboud University.

I’m looking for research internships for 2027 in machine hearing, audio and speech representation learning, or spatial audio. Feel free to reach out by email.

news

Sep 25, 2026	Our paper AudioSphere, the first self-supervised audio model that performs sound event localization and detection from a frozen backbone, was accepted at NeurIPS 2026!
Sep 19, 2026	Two new preprints: Bearings, self-supervised soundfield embeddings that plug into frozen audio encoders, and EquiSELD, efficient training of equivariant sound event localization and detection networks.
Sep 27, 2025	New preprint: WavJEPA, semantic learning for robust audio foundation models on raw waveforms. Code is on GitHub.
Jun 01, 2025	New preprint: GRAM, spatial general-purpose audio representation models, with the NatHEAR and RealSELD benchmarks.
Sep 01, 2024	Started my PhD at the Donders Institute, Radboud University, on machine hearing and representation learning for spatial audio.

selected publications

NeurIPS 2026
AudioSphere: Towards Self-Supervised Spatial Audio Representation Models

Göksenin Yüksel, Marcel Gerven, and Kiki Heijden

In Advances in Neural Information Processing Systems (NeurIPS), 2026

Abs Bib Code

Recent audio embedding models learn general-purpose representations that transfer across clip- and frame-level tasks, yet they operate almost exclusively on single-channel signals and are blind to spatial structure. We introduce AudioSphere, a multi-channel masked autoencoder trained on Ambisonics recordings that reconstructs both spectral content and directional cues via a joint spatiotemporal masking strategy. Alongside the model, we release RealSELD, a standardized benchmark for evaluating frozen backbones on real-world multi-channel sound event localization and detection (SELD). AudioSphere enables, for the first time, SELD with a frozen self-supervised backbone, while matching or exceeding prior general-purpose audio embedding models on the HEAR benchmark.
@inproceedings{yuksel2026audiosphere, title = {AudioSphere: Towards Self-Supervised Spatial Audio Representation Models}, author = {Yüksel, Göksenin and van Gerven, Marcel and van der Heijden, Kiki}, booktitle = {Advances in Neural Information Processing Systems (NeurIPS)}, year = {2026}, }
Under review

SpeechJEPA: Clustering-Free Latent Prediction for Compute-Efficient Speech Representation Learning

Göksenin Yüksel, Marcel Gerven, and Kiki Heijden

Under review at ICASSP 2027

Abs Code

Joint-embedding predictive architectures (JEPAs) were reported to collapse on speech unless anchored by explicit input-level clustering. SpeechJEPA is a clustering-free JEPA that learns speech representations entirely in latent space, without discrete targets, negative samples or auxiliary grounding. A mask-conditioned convolutional predictor regresses several target groups from a single shared context encoding, so the encoder runs once per utterance. Pre-trained on LibriSpeech in 13.8 GPU-hours on one H100, SpeechJEPA beats wav2vec 2.0 with one hour of labels at about 17x fewer training FLOPs, and a 64-GPU-hour model outperforms wav2vec 2.0, HuBERT and WavLM at every labeled-data budget.
Under review
Bearings: Self-Supervised Soundfield Embeddings from First-Order Ambisonics

Göksenin Yüksel, Marcel Gerven, and Kiki Heijden

Under review at ICASSP 2027; arXiv:2609.23152

Abs Bib HTML PDF Code

Audio encoders capture acoustic content well but have no notion of where sounds come from. Bearings learns spatial audio representations without labels by training a masked autoencoder on first-order Ambisonics, with a decoder conditioned on frozen embeddings from single-channel acoustic encoders. The resulting soundfield embeddings can be combined with different frozen acoustic models without retraining either one. Fused with acoustic representations for sound event localization and detection, they raise the location-dependent F-score from below 4 to 50 on TAU-NIGENS 2021 and to 39 on STARSS23.
@article{yuksel2026bearings, title = {Bearings: Self-Supervised Soundfield Embeddings from First-Order Ambisonics}, author = {Yüksel, Göksenin and van Gerven, Marcel and van der Heijden, Kiki}, journal = {Under review at ICASSP 2027; arXiv:2609.23152}, year = {2026}, }
Under review
EquiSELD: Efficient training of equivariant sound event localization and detection networks

Göksenin Yüksel, Marcel Gerven, and Kiki Heijden

Under review at ICASSP 2027; arXiv:2609.23156

Abs Bib HTML PDF Code

First-order Ambisonics signals have an exact O(3) symmetry: rotating or reflecting the signal changes the direction of arrival of the sound sources but not the sources themselves. EquiSELD processes Ambisonics as paired streams of O(3)-invariant scalars and equivariant intensity vectors, producing an invariant activity magnitude and an equivariant direction of arrival with a Multi-ACCDOA readout. It outperforms prior equivariant approaches with far fewer parameters and less training time, outperforms non-equivariant SELD networks on simulated scenes, and is competitive on real-world recordings.
@article{yuksel2026equiseld, title = {EquiSELD: Efficient training of equivariant sound event localization and detection networks}, author = {Yüksel, Göksenin and van Gerven, Marcel and van der Heijden, Kiki}, journal = {Under review at ICASSP 2027; arXiv:2609.23156}, year = {2026}, }
Under review
WavJEPA: Semantic learning unlocks robust audio foundation models for raw waveforms

Göksenin Yüksel, Pierre Guetschel, Michael Tangermann, Marcel Gerven, and Kiki Heijden

Under review at IEEE/ACM Transactions on Audio, Speech, and Language Processing; arXiv:2509.23238

Abs Bib HTML PDF Code Website

Learning audio representations from raw waveforms overcomes key limitations of spectrogram-based audio representation learning, such as the latency of spectrogram computation and the loss of phase information. We propose WavJEPA, a waveform-based version of the Joint-Embedding Predictive Architecture that learns high-level semantic representations, substantially outperforming state-of-the-art time-domain audio foundation models while requiring considerably fewer computational resources. We also present WavJEPA-Nat, a multi-channel extension trained on simulated naturalistic scenes that is highly robust to reverberation and noise.
@article{yuksel2025wavjepa, title = {WavJEPA: Semantic learning unlocks robust audio foundation models for raw waveforms}, author = {Yüksel, Göksenin and Guetschel, Pierre and Tangermann, Michael and van Gerven, Marcel and van der Heijden, Kiki}, journal = {Under review at IEEE/ACM Transactions on Audio, Speech, and Language Processing; arXiv:2509.23238}, year = {2025}, }
arXiv
GRAM: Spatial general-purpose audio representations

Göksenin Yüksel, Marcel Gerven, and Kiki Heijden

arXiv preprint arXiv:2506.00934, 2025

Abs Bib HTML PDF Code

Most audio foundation models ignore the spatial structure of real-world sound scenes. We propose GRAM, a general-purpose real-world audio model that uses a multi-channel masked autoencoder to efficiently learn spatial audio representations. We evaluate GRAM and other audio foundation models on high-quality simulations of naturalistic spatial acoustic environments and on real-world recordings, and release two complementary benchmark task suites: NatHEAR and RealSELD.
@article{yuksel2025gram, title = {GRAM: Spatial general-purpose audio representations}, author = {Yüksel, Göksenin and van Gerven, Marcel and van der Heijden, Kiki}, journal = {arXiv preprint arXiv:2506.00934}, year = {2025}, }
MIDL
Style Randomization Improves the Robustness of Breast Density Estimation in MR Images

Göksenin Yüksel, Koen Eppenhof, Jaap Kroes, and Marcel Worring

In Proceedings of the 7th International Conference on Medical Imaging with Deep Learning (MIDL), 2024

Abs Bib HTML PDF

Breast density, a crucial risk factor for future breast cancer development, is defined by the ratio of fat to fibro-glandular tissue (FGT) in the breast. Accurate breast and FGT segmentation is essential for robust density estimation. We propose a novel data augmentation technique, motivated by style randomization, to further exploit a multi-modal training setup with and without fat suppression. Networks trained with the proposed augmentation are resilient to variations in fat content and show improved robustness compared to training with multi-modal data alone.
@inproceedings{yuksel2024style, title = {Style Randomization Improves the Robustness of Breast Density Estimation in MR Images}, author = {Yüksel, Göksenin and Eppenhof, Koen and Kroes, Jaap and Worring, Marcel}, booktitle = {Proceedings of the 7th International Conference on Medical Imaging with Deep Learning (MIDL)}, series = {Proceedings of Machine Learning Research}, volume = {250}, pages = {1841--1850}, year = {2024}, publisher = {PMLR}, }
MLRC
Reproducibility Study of "Quantifying Societal Bias Amplification in Image Captioning"

Farrukh Baratov^*, Göksenin Yüksel^*, Darie Petcu^*, and Jan Bakker^*

In NeurIPS 2023 Main Hall Poster, ML Reproducibility Challenge 2022

Abs Bib HTML PDF

We study the reproducibility of "Quantifying Societal Bias Amplification in Image Captioning" by Hirota et al., which proposes the LIC metric for measuring bias amplification. Our reproduced results support all four main claims of the original paper. We also extend the work by evaluating LIC for age bias, finding that only a subset of the models amplifies age bias.
@inproceedings{baratov2023reproducibility, title = {Reproducibility Study of "Quantifying Societal Bias Amplification in Image Captioning"}, author = {Baratov, Farrukh and Yüksel, Göksenin and Petcu, Darie and Bakker, Jan}, booktitle = {NeurIPS 2023 Main Hall Poster, ML Reproducibility Challenge 2022}, year = {2023}, }