Paulin Melatagia Yonta

Le Mans Université · Cameroon

Unclaimed · AI-structured profileORCID OpenAlex

Le Mans Université speech researcher applying heterogeneous graph neural networks to low-resource language Yemba

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—

Not yet scored: a complete publication-level output assessment is not available for this profile. This axis would measure total indexed works and their distribution over time; only partial year counts and a thin citation record (h-index 4, i10-index 0).

Impact—

Not yet scored: the citation record is too sparse to assess influence. This axis would measure citations received and downstream effect on the field; what is on file (h-index 4, i10-index 0) does not yet support a reliable score.

Focus0/100

Focus scores 0, the lowest concentration band. Work splits evenly across two fields — speech processing / multimodal representation learning (0.5) and speech processing / computational linguistics (0.5) — and the topic list spans five areas: speech recognition and synthesis (15), neural networks and applications (10), speech and audio processing (9), topic modeling (8), and natural language processing techniques (7). The profile reads as methodological breadth serving low-resource speech rather than a single narrow line.

Collaboration22/100

Collaboration scores 22, the weakest measured axis: only 2 distinct co-authors appear in the recent OpenAlex window. The recurring partners are Yannick Yomie Nzeuhang and Marie Tahon. On the two catalog papers Yonta is a non-corresponding author — middle position on one, last on the other — fitting a contributing or senior-supporting role inside a small working group rather than leadership of a large consortium.

Activity93/100

Activity scores 93, the strongest axis in the radar. After modest annual output from 2005 to 2019 (mostly 1-2 indexed works per year, with a first peak of 8 in 2020), production accelerated sharply: 8 works in 2024, 10 in 2025, and 5 already in 2026, totaling 23 works in the recent three-year window. This points to an intense, ongoing publishing phase around the current low-resource speech program.

Reproducibility55/100

Reproducibility scores 55, computed by Evolve across 2 assessed papers, both rated medium readiness. The reviews identify concrete gaps: one paper lacks a key acoustic-model baseline and its Yemba gain is not statistically significant; the other reports single-run evaluation without frugal baselines. Methods are graph-based and described in usable detail, but missing controls currently keep replication from being straightforward.

Research summary

Paulin Melatagia Yonta is a researcher at Le Mans Université working on speech processing for low-resource languages; this profile is name-resolved from OpenAlex bibliographic data linked to ORCID 0000-0003-3479-2627. The current research line centers on heterogeneous graph neural networks (HGNN) for cross-modal transfer between text and audio: two 2026 arXiv preprints, both in the Evolve catalog, use HGNN-based knowledge transfer and link prediction to strengthen speech representation learning for Yemba, a low-resource language with little annotated data. One targets isolated word recognition under low-resource settings; the other performs word-level speech–text alignment with acoustic word embeddings, reporting gains over SAMU-XLSR. The broader topic profile spans speech recognition and synthesis (15 works), neural networks and applications (10), speech and audio processing (9), topic modeling (8), and natural language processing techniques (7). Output is concentrated in a recent surge: 23 works over the last three years, with 8 indexed in 2024, 10 in 2025, and 5 so far in 2026, against annual counts of roughly 1-2 from 2005 to 2019. The citation record is thin (h-index 4, i10-index 0), so output and impact remain unscored. Collaboration is compact, with 2 distinct recent co-authors, notably Yannick Yomie Nzeuhang and Marie Tahon, and Evolve rates reproducibility medium on both reviewed papers.

Collaboration pattern

Collaboration is compact: the recent OpenAlex window lists only 2 distinct co-authors, and the collaboration radar score is 22. The recurring partners are Yannick Yomie Nzeuhang and Marie Tahon. On both catalog papers Yonta is a non-corresponding author — middle position on one, last on the other — pointing to a contributing or senior-supporting role inside a small working group rather than leadership of a large consortium. The tight author set makes the 2024-2026 output surge notably two-person-driven.

Research keywords

speech representation learningheterogeneous graph neural networkscross-modal alignmentcross-modal knowledge transferlow-resource languagesacoustic word embeddingsspeech recognition and synthesisspeech and audio processingtopic modelingnatural language processingYemba

Highlights

Frugal HGNN link prediction for word-level speech–text alignment in Yemba
The 2026 preprint proposes a frugal heterogeneous graph neural network that casts word-level speech–text alignment as link prediction to enhance speech representations with acoustic word embeddings. Evolve's review judges the reported Yemba gains over SAMU-XLSR as looking real, but flags a single-run evaluation and missing frugal baselines as limits; reproducibility is rated medium.
Cross-modal knowledge transfer with HGNN for low-resource Yemba word recognition
The companion 2026 preprint applies HGNN-based explicit text-to-audio knowledge transfer to speech representation learning under low-resource settings, evaluated on isolated word recognition for Yemba. Evolve finds gains on low-resource word recognition but notes a missing key acoustic-model baseline and a Yemba gain that is not statistically significant; reproducibility is rated medium.
Sharply accelerating publication activity since 2024
After years of roughly 1-2 indexed works annually from 2005 to 2019 (with a first peak of 8 in 2020), output jumped to 8 works in 2024, 10 in 2025, and 5 already in 2026 — 23 works in the recent three-year window, driving an activity score of 93.

Timeline

2005Earliest year in the OpenAlex record, with 1 indexed work.
2020An early peak of 8 indexed works, the highest annual count before the current surge.
20248 indexed works, marking the start of the current acceleration.
202510 indexed works, the highest annual output on record.
20265 works indexed so far, plus two HGNN-based Yemba preprints entering the Evolve catalog.

Papers in the Evolve catalog

Frequent co-authors

Yannick Yomie Nzeuhang×2 Marie Tahon×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.

Bibliometrics come from OpenAlex only; the citation record on file is thin (h-index 4, i10-index 0), so the output and impact axes are unscored.
Evolve's qualitative verdicts cover only the 2 catalog papers (both 2026 arXiv preprints) and may not represent the full body of work.
The reproducibility score (mean 55) rests on just 2 assessed papers — a small sample.
Profile name-resolved from public bibliographic data linked to ORCID 0000-0003-3479-2627; affiliation as indexed (Le Mans Université, country code CM).