LaVonne Roberts

Société mycologique de France

Unclaimed · AI-structured profileORCID OpenAlex

Clinical speech AI researcher: per-aetiology dysarthria severity modelling and PEFT-based per-patient ASR adaptation

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

—

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—

This axis would measure overall publication volume, yearly output trends and authorship roles across the full record. The current snapshot provides no such metrics — only the two reviewed papers exist here — so no score can be reported yet.

Impact—

Impact would capture citation counts, citations per paper and scholarly influence over time. No citation data is present in this snapshot — Evolve's qualitative verdicts exist, but no quantitative uptake metrics — so this axis stays unscored.

Focus0/100

The radar focus score is 0 in this snapshot, even though the field distribution splits evenly — 0.5 and 0.5 — between two nearly identical labels, 'speech technology (clinical speech AI)' and 'speech recognition / speech technology (machine learning for clinical speech)'. Both catalogued papers target dysarthric speech, so topically the record is tightly concentrated; the zero likely reflects how the scorer treats the two-label split on a two-paper corpus rather than genuine breadth.

Collaboration28/100

The collaboration score is 28, based on 3 distinct coauthors in the raw data, indicating a small-team working mode. The recurring partners are Bernard Muller, Antonio Armando Ortiz Barrañón and László Tóth. On both catalogued papers Roberts occupies the last-author position without a corresponding-author flag, consistent with a senior or supervisory role in a compact group rather than large-consortium collaboration.

Activity—

Activity would measure publishing cadence — recency, frequency and consistency of outputs over time. Beyond the two 2026 papers, no longitudinal production data is available in this snapshot, so the axis cannot be scored yet.

Reproducibility55/100

The reproducibility score is 55, computed as the mean over 2 papers, both rated 'medium' readiness. Evolve's verdicts describe careful, well-controlled designs — ablation-style PEFT comparisons on two ASR bases, per-aetiology versus pooled training — but also single-run experiments, SAP-dominated severity test sets, pseudo-label agreement of only κ≈0.07, and an n=1 speaker study, all of which cap how far results can be independently reproduced or generalized.

Research summary

LaVonne Roberts (listed affiliation: Société mycologique de France; ORCID 0009-0008-2883-3764) works on machine learning for clinical speech, with both catalogued outputs — two 2026 arXiv papers — sitting squarely in dysarthric speech processing. The first, with Roberts in the last-author position, introduces per-aetiology contrastive severity embeddings with phonological pseudo-labelling for multilingual dysarthric speech built on HuBERT; Evolve's review finds per-aetiology models beat pooled training for cerebral palsy, Parkinson's disease and ALS severity by +23–40% relative F1, while noting single-run experiments, SAP-dominated test sets, and weak pseudo-label agreement (κ≈0.07). The second is a single-speaker case study of parameter-efficient fine-tuning for per-patient dysarthric ASR across two ASR bases, where LoRA ties DoRA and beats 4-bit QLoRA and four other PEFT variants; the n=1 design limits generality. Radar scores reflect the sparse record: collaboration 28 with 3 distinct coauthors (recurring partners Bernard Muller, Antonio Armando Ortiz Barrañón and László Tóth), reproducibility 55 averaged over 2 papers, and a focus score of 0 despite an even 0.5/0.5 split between two near-identical speech-technology labels. Output, impact and activity axes are unscored for lack of data, making this an early-stage profile built on two carefully controlled but small-scale studies.

Collaboration pattern

Roberts works in a compact team: the raw record lists only 3 distinct coauthors, and the same three names — Bernard Muller, Antonio Armando Ortiz Barrañón and László Tóth — recur across the catalog. Both papers place Roberts last without a corresponding-author flag, suggesting a senior or advisory role. The collaboration radar score of 28 confirms a low-breadth, high-stability pattern; no large-consortium or cross-institution projects appear in this snapshot.

Research keywords

Dysarthria severity classificationAetiology-specific modelsCerebral palsyParkinson's diseaseALSHuBERTDysarthric speech recognitionParameter-efficient fine-tuningLoRADoRAQLoRAPer-patient adaptation

Highlights

Per-aetiology severity models gain +23–40% relative F1 over pooled training
In the 2026 multilingual dysarthric speech paper, contrastive severity embeddings trained per aetiology outperform pooled models for cerebral palsy, Parkinson's disease and ALS. Evolve calls the work well-controlled but flags single-run experiments, SAP-dominated evaluation and pseudo-label agreement of κ≈0.07. The approach is built on HuBERT representations.
Practical PEFT guidance: LoRA ties DoRA and beats 4-bit QLoRA for per-patient dysarthric ASR
A single-speaker case study compares parameter-efficient fine-tuning variants across two ASR bases: LoRA matches DoRA and outperforms 4-bit QLoRA along with four other variants. Evolve rates the work careful, but the n=1 design limits generality of the ranking.
Phonological pseudo-labelling pipeline with honest error accounting
The multilingual severity paper introduces phonological pseudo-labelling to expand the training signal; Evolve notes pseudo-label agreement is weak (κ≈0.07), yet in the controlled comparison per-aetiology training still wins across cerebral palsy, Parkinson's disease and ALS. Given the single-run design, reported gains should be read as preliminary.

Timeline

2026Publishes 'Per-Aetiology Contrastive Severity Embeddings with Phonological Pseudo-Labelling for Multilingual Dysarthric Speech' on arXiv, in the last-author position.
2026Publishes 'Choosing a PEFT Variant for Per-Patient Dysarthric ASR: A Single-Speaker Case Study on Two ASR Bases' on arXiv, again as last author.
2026Both catalogued outputs land in the same year, concentrating the profile on machine learning for clinical dysarthric speech with three recurring coauthors.

Papers in the Evolve catalog

Frequent co-authors

Bernard Muller×2 Antonio Armando Ortiz Barrañón×1 László Tóth×1

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 come from a sparse public bibliographic snapshot; output, impact and activity axes lack underlying data and are unscored rather than estimated.
The profile rests on two 2026 arXiv papers reviewed by Evolve; no citation counts, venue acceptance records or funding data are included.
The listed affiliation, Société mycologique de France, does not match the clinical speech AI research fields; this profile is name-resolved from public bibliographic data (ORCID 0009-0008-2883-3764), so the affiliation may reflect a registry mismatch.
Reported effects (+23–40% relative F1; PEFT rankings from an n=1 speaker) come from single-run, small-scale designs per Evolve verdicts and should be treated as preliminary.