Goksenin Yuksel

Donders Institute for Brain, Cognition and Behaviour · Netherlands

Unclaimed · AI-structured profileOpenAlex

Machine learning for spatial audio: equivariant SELD networks and self-supervised soundfield embeddings

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-10-09) and cover the author's entire public record, not just the papers listed here. Reproducibility signal comes from Evolve's own reviews.

Research profile

Output—

Output is not scored: the h-index and i10-index are recorded as 0 and total publication counts are incomplete for this OpenAlex identity. This axis would measure cumulative publication volume across venues and preprint servers over time.

Impact—

Impact is not scored: citation data are not available in the provided record, and both reviewed papers are recent arXiv preprints. This axis would measure citations, uptake, and scholarly influence accumulated after publication.

Focus0/100

The focus score of 0 comes from an even split in the recorded field distribution: 0.5 for machine learning for spatial audio / audio signal processing and 0.5 for machine learning for spatial audio (sound event localization and detection). Both labels describe essentially the same spatial-audio ML theme seen in the reviewed papers, so the low score reflects label granularity more than real dispersion; OpenAlex topics additionally span speech, music, topic modeling, and information retrieval.

Collaboration22/100

The collaboration score of 22 reflects a small recorded network rather than a judged quality of teamwork: OpenAlex lists only 2 distinct coauthors, and the recurring collaborators are Marcel van Gerven and Kiki van der Heijden. Both catalog papers are first-author, non-corresponding works. A compact, stable core team around spatial-audio machine learning suggests focused early-stage collaboration; multi-institution or large-consortium patterns are not visible in the current record.

Activity61/100

The activity score of 61 derives from raw OpenAlex counts: 7 works within the recent three-year window, split as 4 outputs in 2025 and 3 in 2026. This is a steady recent publishing cadence for an early-career record, and two of the 2026 items are first-author arXiv preprints reviewed in the Evolve catalog. The count covers the full OpenAlex record, not only the reviewed papers.

Reproducibility100/100

Reproducibility scores 100, the maximum, computed as the mean readiness over 2 Evolve-reviewed papers, both rated high. EquiSELD reports an explicit O(3)-versus-SO(3) ablation and states where wins sit within noise, while Bearings reports ablated SELD gains (F20° from under 4 to 50.4 on TAU) and quantifies its lead over hand-crafted statistics. Transparent ablations and honest boundary reporting support the score, though it rests on a two-paper sample.

Research summary

Goksenin Yuksel is a researcher at the Donders Institute for Brain, Cognition and Behaviour in the Netherlands working on machine learning for spatial audio. The Evolve catalog holds two first-author arXiv preprints, both from 2026 and both rated high for reproducibility. Bearings introduces self-supervised soundfield embeddings from first-order Ambisonics, claimed to be the first such embeddings pluggable into frozen encoders; ablated SELD gains are solid, with the F20° metric rising from under 4 to 50.4 on TAU, though the lead over hand-crafted statistics is moderate. EquiSELD trains exactly O(3)-equivariant attention networks for sound event localization and detection, winning on TAU2021, while its own O(3)-versus-SO(3) ablation shows no clear gain and STARSS23 wins lie within noise — a candidly reported negative result. Bibliometric coverage is early-stage: OpenAlex records 7 works over the recent three-year window (4 in 2025, 3 in 2026), just 2 distinct coauthors, and h- and i10-index values of 0, which is why the output and impact dimensions are left unscored. The recorded field distribution splits evenly between spatial-audio signal processing and SELD, and OpenAlex topic assignments extend to speech, music, topic modeling, and information retrieval. Frequent coauthors are Marcel van Gerven and Kiki van der Heijden. This profile is name-resolved from public bibliographic data; no ORCID is linked.

Collaboration pattern

The visible pattern is a compact core team: OpenAlex records only 2 distinct coauthors, matching the low collaboration score of 22, and the recurring collaborators are Marcel van Gerven and Kiki van der Heijden. Both reviewed papers are first-author and non-corresponding. The work sits at the Donders Institute for Brain, Cognition and Behaviour in the Netherlands. No multi-institution or large-consortium collaborations appear in the provided record.

Research keywords

Sound event localization and detection (SELD)Spatial audio machine learningFirst-order AmbisonicsEquivariant neural networksO(3) and SO(3) equivarianceEquivariant attentionSelf-supervised learningMasked autoencodersSoundfield embeddingsAudio signal processingRepresentation learning for audioSELD benchmarks (TAU, STARSS)

Highlights

Self-supervised soundfield embeddings that plug into frozen encoders
Bearings claims the first self-supervised soundfield embeddings from first-order Ambisonics that can be plugged into frozen encoders. Ablated SELD gains are solid, with the F20° metric rising from under 4 to 50.4 on TAU, though the lead over hand-crafted statistics is moderate.
Exactly O(3)-equivariant SELD network with efficient training
EquiSELD is an exactly O(3)-equivariant attention network for sound event localization and detection that leads on TAU2021, framed around efficient training of equivariant SELD architectures.
Candid negative results in the equivariance ablation
The same paper's O(3)-versus-SO(3) ablation shows no clear gain from the larger symmetry group, and its STARSS23 wins lie within noise — an unusually transparent boundary report for benchmark-driven SELD work.

Timeline

2025OpenAlex records 4 works in 2025, the earliest year in the provided activity window.
2026First-author arXiv preprint Bearings introduces self-supervised soundfield embeddings from first-order Ambisonics.
2026First-author arXiv preprint EquiSELD presents exactly O(3)-equivariant sound event localization and detection networks.
2026OpenAlex records 3 works in 2026, bringing the recent three-year total to 7; both catalog papers are rated high for reproducibility (mean 100 over 2 papers).

Papers in the Evolve catalog

2026-09-19first author
EquiSELD: Efficient training of equivariant sound event localization and detection networks
2026-09-19first author
Bearings: Self-Supervised Soundfield Embeddings from First-Order Ambisonics

Frequent co-authors

Marcel van Gerven×2 Kiki van der Heijden×2

About this profile

This profile is AI-structured by Evolve.Pub from public bibliographic data (OpenAlex) and Evolve's own paper reviews. It is not written or endorsed by the researcher unless the profile is claimed. Radar dimensions are computed from real bibliometrics with fixed formulas — raw figures are shown alongside; missing dimensions mean insufficient data, never zero. Errors? Claim the profile to correct or hide it.

Bibliometric figures come from OpenAlex only; no ORCID is linked, so the profile is name-resolved from public bibliographic data and could merge or split identities.
Both reviewed papers are 2026 arXiv preprints; peer-reviewed venue status is not recorded.
Output and impact dimensions are unscored because h- and i10-index are recorded as 0 and citation data are absent — an early bibliometric footprint, not a measured citation record.
Highlights and the reproducibility score rest on a small sample: 2 Evolve-reviewed papers with a mean readiness of 100.