Aalto University researcher building interpretable native-reference geometry for L2 pronunciation assessment
2
papers here
2
reviewed by Evolve
2
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—
No output-volume data is available yet: total publication counts, venue types, and per-year productivity cannot be computed from the current record, which shows only two 2026 arXiv preprints. This axis would measure cumulative and yearly publishing output across the full bibliographic record.
Impact—
Citation impact data is not yet available — this profile holds no citation counts, h-index, or field-normalized rates. The axis would measure how often the researcher's publications are cited relative to field and year baselines, which is premature for 2026 preprints.
Focus0/100
The focus axis scores 0 on the 0-100 scale, computed from the field distribution in the catalog: two labeled fields — speech processing / spoken language assessment (computational phonetics) and speech processing / computational linguistics (computer-assisted language learning) — split evenly at 0.5 each. In substance the spread is narrow, since both halves concern speech processing for L2 pronunciation assessment, but the exact 50/50 split across two field labels yields the minimal focus score.
Collaboration39/100
The collaboration axis scores 39 on the 0-100 scale, derived from a raw count of 6 distinct co-authors across the two catalogued papers. The recurring collaborators are Nhan Phan, Mikko Kurimo, and Chenxiao Wang. This points to a compact, stable working group of the kind typical of a single lab, with limited breadth of external partnerships so far; the moderate score reflects small-team science rather than absent collaboration.
Activity—
Activity data is insufficient: the visible record spans a single year (2026) with two preprints, so publishing cadence, recency trends, and long-run continuity cannot be scored. This axis would track output tempo and sustained productivity across the researcher's full career record.
Reproducibility78/100
Reproducibility is the strongest scored axis at 78, the mean readiness rating across 2 papers: high for the phone-class geometry preprint and medium for the coordinate-geometry preprint. Evolve's reviews support the rating's credibility while noting limits — one paper is tuned on train/dev data only with no held-out test set — so the score reflects methodological transparency and artifact readiness more than externally validated replication.
Research summary
Tina Raissi is a speech-processing researcher at Aalto University working on second-language (L2) pronunciation assessment. The Evolve catalog records two 2026 arXiv preprints, with Raissi as first author on both. The papers express one coherent research program: extract phone-class embeddings from self-supervised speech models such as WavLM, anchor them to a native-reference geometry, and use distances — organized through singular value decomposition (SVD) — as an interpretable, training-free measure of pronunciation deviation. The coordinate-geometry paper reports correlations with CEFR-level English speaking proficiency of roughly ρ ≈ −0.5; the phone-class geometry paper reports ρ −0.53/−0.34 against L2 proficiency, while acknowledging that much of the signal overlaps basic fluency features. Evolve's verdicts describe the work as solid, sound, and honestly caveated, while flagging that tuning used train/dev data only, with no held-out test set. Reproducibility readiness is rated high for one paper and medium for the other, a mean of 78 across 2 papers. The recorded collaboration footprint is compact — 6 distinct co-authors, with Nhan Phan, Mikko Kurimo, and Chenxiao Wang recurring. This is an early-stage profile: citation impact, output volume, and multi-year activity metrics are not yet available.
Collaboration pattern
Raissi's recorded collaboration footprint is compact: 6 distinct co-authors across the two catalogued preprints, yielding a collaboration score of 39 on Evolve's 0-100 scale. Three names recur most often — Nhan Phan, Mikko Kurimo, and Chenxiao Wang — appearing across the catalogued work. The pattern suggests a small, stable research group at Aalto University rather than a wide external network; no international co-authorship pattern can be inferred from the available data.
Research keywords
Highlights
Interpretable phone-class geometry correlates with L2 proficiency (ρ −0.53/−0.34)
Training-free SVD coordinate geometry tracks CEFR speaking proficiency (ρ ≈ −0.5)
One consistent methodology: native-reference, training-free geometry across both papers
Both catalogued preprints pursue the same program — measuring L2 pronunciation deviation as distance in a geometry anchored to native-reference phone-class embeddings from self-supervised speech models, structured by SVD rather than learned output heads. Evolve's shared caveat applies across the line: part of the discrimination overlaps basic fluency features.
Timeline
- 2026First-author arXiv preprint introduces a training-free native-reference coordinate geometry built on self-supervised speech models, tracking L2 English proficiency against CEFR speaking levels at ρ ≈ −0.5.
- 2026A companion first-author arXiv preprint proposes an interpretable phone-class geometry, with distances negatively correlating with L2 proficiency at ρ −0.53/−0.34 and high reproducibility readiness.
- 2026Catalog snapshot for the year: two first-author preprints in speech processing for language learning, a reproducibility mean of 78 across 2 papers, and 6 distinct co-authors.
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 are derived from public bibliographic data (OpenAlex-based); no ORCID or country metadata was available, so identity is name-resolved.
- Output, impact, and activity dimensions are unscored due to insufficient data; no citation counts or yearly output figures are quoted in this profile.
- The evidence base is two 2026 arXiv preprints in the Evolve catalog; peer-reviewed venues, funding, and awards are not covered.
- The reproducibility score (78) is a mean over only 2 papers; both works are arXiv-stage and lack externally validated replication.