Audio AI researcher — fairness auditing, adversarial robustness, and turn-taking evaluation for speech dialogue models
3
papers here
3
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-02) 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: the snapshot has no verified publication counts, venue records, or yearly output totals for this name. This axis would measure total scholarly output and its growth over the researcher's career.
Impact—
Not yet scored: no citation counts, h-index, or venue-level impact data are linked to this profile. This axis would measure citation-based influence, such as citations per paper and overall scholarly reach, once such records exist.
Focus0/100
Not yet scored: structured field and keyword classifications across the full corpus are unavailable, so topical concentration cannot be quantified. The three reviewed titles suggest a coherent audio-AI theme, but that is not yet a measured focus score.
Collaboration14/100
Scored 14 from bibliometric data showing only 1 distinct co-author across the cataloged papers. The network is a compact two-person cluster: Kian Shamsaie is the sole recurring collaborator, and Modarressi takes the last author position on all three 2026 arXiv preprints without being marked corresponding. This is consistent with senior supervision inside a small, stable team rather than broad multi-group collaboration.
Activity—
Not yet scored: yearly publication counts and recent-activity windows are absent from the snapshot. This axis would measure publishing cadence and continuity in recent years, which three 2026 arXiv entries alone cannot establish.
Reproducibility55/100
Not yet scored: none of the three reviewed papers carries a reproducibility-readiness assessment or code-release record. This axis would measure artifact availability, documentation quality, and whether reported experiments can be independently re-run.
Research summary
Iman Modarressi is an early-stage profile in Evolve's catalog, reconstructed from public bibliographic data and three 2026 arXiv preprints, each with Modarressi in the last (senior) author position. The work concentrates on trustworthy audio and speech AI along three complementary lines. The first audits demographic fairness in audio understanding models, arguing that representation entanglement can lower-bound disparity and using that link both to measure and to repair unfair behavior. The second targets adversarial robustness of full-duplex speech-to-speech dialogue models, introducing psychoacoustically aligned latent smoothing so that perturbation robustness respects human auditory perception. The third rethinks evaluation methodology for full-duplex spoken dialogue systems, proposing an intent-conditioned view of turn-taking in which neither silence nor overlap is automatically a failure. Together these papers indicate a research agenda spanning fairness auditing, adversarial robustness, and interaction evaluation, all centered on speech interfaces rather than text-only models. Quantitative context remains thin: the current snapshot scores only the collaboration dimension (14 of 100), built on a single distinct co-author, Kian Shamsaie, who recurs across the catalog. Institution, country, ORCID, citation counts, and venue records are not yet linked, so this profile should be read as a focused, recently active last author whose bibliometric footprint is still accumulating rather than as a fully measured career record.
Collaboration pattern
The observed pattern is a tight two-person collaboration: exactly 1 distinct co-author appears across the cataloged papers, and Kian Shamsaie is the sole recurring collaborator. In all three 2026 arXiv papers Modarressi sits in the last author position without being marked corresponding, consistent with a senior-supervision role inside a compact team. No institutional or cross-group collaborations are visible in the current snapshot.
Research keywords
Highlights
Entanglement as a lower bound on demographic disparity in audio models
Last-authored 2026 arXiv preprint proposing to audit demographic fairness in audio understanding models via representation entanglement, and to repair the unfair behavior that the entanglement-disparity link exposes.
Psychoacoustically aligned latent smoothing for speech-to-speech robustness
Last-authored 2026 arXiv preprint introducing latent smoothing aligned with psychoacoustics to improve the adversarial robustness of full-duplex speech-to-speech dialogue models.
Intent-conditioned evaluation of turn-taking in spoken dialogue
Last-authored 2026 arXiv preprint reframing turn-taking evaluation for full-duplex spoken dialogue models, arguing that neither silence nor overlap is inherently a failure and conditioning judgments on dialogue intent.
Timeline
- 2026Last-authored arXiv preprint auditing and repairing demographic fairness in audio understanding models via entanglement lower bounds.
- 2026Last-authored arXiv preprint on psychoacoustically aligned latent smoothing for adversarial robustness of full-duplex speech-to-speech dialogue models.
- 2026Last-authored arXiv preprint proposing intent-conditioned evaluation of turn-taking in full-duplex spoken dialogue models.
Papers in the Evolve catalog
When Entanglement Lower-Bounds Disparity: Auditing and Repairing Demographic Fairness in Audio Understanding Models
Psychoacoustically Aligned Latent Smoothing for Adversarial Robustness of Full-Duplex Speech-to-Speech Dialogue Models
Neither Silence nor Overlap Is Failure: Intent-Conditioned Evaluation of Turn-Taking in Full-Duplex Spoken Dialogue Models
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.
- Radar scores come from a single bibliometric snapshot; only the collaboration axis had supporting data (1 distinct co-author), so cross-source verification is pending.
- All cataloged papers are 2026 arXiv preprints; peer-review status and final venues are not captured.
- Institution, country, and ORCID are unlinked; the profile is name-resolved from public bibliographic data, and same-name researchers cannot be fully excluded.
- No citation counts, h-index, venue metrics, or code-release records were provided, so impact and reproducibility remain unmeasured.