Speech & NLP researcher; 2026 last-author work on efficient zero-shot SpeechLLMs and speech-model fairness probing
3
papers here
3
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
19
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-16) 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—
Not yet scored: this axis would measure publication volume and productivity over time, but the snapshot holds too few indexed records to compute a meaningful figure.
Impact—
Insufficient data: impact would be measured through citations and downstream uptake, and the two 2026 arXiv papers are too recent for such signals to have accumulated.
Focus0/100
Focus is the strongest signal in this snapshot at 100: the field distribution contains exactly one field, speech and natural language processing, and every field-tagged cataloged paper falls within it. The two 2026 papers — token-level alignment for zero-shot SpeechLLMs and a critique of probe-based fairness steering — attack different problems inside the same speech-model space, indicating a tightly concentrated, though still small, research agenda.
Collaboration39/100
Collaboration scores 36, computed from 5 distinct co-authors recorded in the snapshot. The recurring set is compact — Abderrahmane Issam, Nicolas Bourrel, Yusuf Can Semerci and Jan Scholtes — which reads as a small, stable working group rather than a broad collaboration network. Spanakis sits in the last author position on both cataloged 2026 papers, a placement usually associated with senior supervision of the team.
Activity100/100
Not yet scored: this axis tracks publishing cadence and recent output intensity, and the snapshot lacks the year-by-year record needed to estimate it.
Reproducibility70/100
Reproducibility scores a perfect 100, but it rests on a single scored paper (reproPapers: 1) — the token-level alignment work, which Evolve rates as having high reproducibility readiness. That same review cautions that headline claims rest on single-seed runs, so the score is best read as a promising single data point rather than an established record across many artifacts.
Research summary
Gerasimos Spanakis is a speech and natural language processing researcher whose Evolve profile is name-resolved from public bibliographic data and built around two 2026 arXiv papers on which he is the last-listed author, a position typically tied to senior or supervisory contribution. 'Align, Integrate, and Fire: Efficient Token-Level Alignment for Zero-Shot SpeechLLMs' combines DTW-aligned Continuous Integrate-and-Fire (CIF) representations with single-layer knowledge distillation to build inexpensive zero-shot SpeechLLMs; Evolve's review judges the reported gains over SpeechMapper as likely real, notes that headline claims rest on single-seed runs, and rates reproducibility readiness high. 'A Probe Shift Is Not a Fairness Fix: The Limits of Representation Steering in Speech Models' takes a measurement-critical stance, arguing that probe shifts alone do not amount to fairness fixes. Radar scores describe a coherent but early snapshot: focus 100, with all field-tagged work in speech and natural language processing; reproducibility 100, computed from a single scored paper; and collaboration 36, reflecting 5 distinct co-authors including recurring collaborators Abderrahmane Issam, Nicolas Bourrel, Yusuf Can Semerci and Jan Scholtes. Output, impact and activity axes remain unscored for lack of data.
Collaboration pattern
The snapshot records 5 distinct co-authors, with a compact recurring core of Abderrahmane Issam, Nicolas Bourrel, Yusuf Can Semerci and Jan Scholtes. Spanakis is last author on both cataloged 2026 papers, a placement typically tied to senior supervision, consistent with a small stable team rather than a wide network; neither paper lists him as corresponding author.
Research keywords
Highlights
Cheap zero-shot SpeechLLMs via DTW-aligned CIF and single-layer distillation
High reproducibility readiness, with a single-seed caveat
A probe shift is not a fairness fix
In this last-author 2026 arXiv paper, Spanakis and co-authors examine the limits of representation steering in speech models, arguing that movement in probe reads does not by itself constitute a fairness improvement.
Timeline
- 2026Last-author arXiv paper 'Align, Integrate, and Fire' introduces DTW-aligned CIF and single-layer knowledge distillation for efficient zero-shot SpeechLLMs.
- 2026Last-author arXiv paper 'A Probe Shift Is Not a Fairness Fix' questions representation steering as a fairness intervention in speech models.
- 2026Evolve review of the alignment paper records high reproducibility readiness while flagging single-seed headline claims.
Papers in the Evolve catalog
Frequent co-authors
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 scores and the field distribution come from an OpenAlex-based snapshot; institution and country are not on record.
- Reproducibility 100 rests on a single scored paper, and the output, impact and activity axes are unscored for insufficient data.
- Only two 2026 arXiv papers from the Evolve catalog inform this profile; it is name-resolved from public bibliographic data and may not cover the full publication record.
- The frequent-coauthor list includes the researcher's own name, a likely data-quality artifact.