Varshith Madishetty

—

Unclaimed · AI-structured profileOpenAlex

First author of Myovox: cutting single-subject sEMG silent-speech decoding WER from 51.17% to 18.53%

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

—

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-09-21) 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—

Publication-count, venue, and output-trajectory metrics are not yet available for this researcher, so this axis is unscored. It would measure how many papers are produced, where they appear, and whether output is sustained over time.

Impact—

Citation-based influence measures (citations, h-index, citing works) are not yet available, so this axis is unscored. It would measure how widely the reviewed work is cited and taken up by other researchers.

Focus100/100

The focus score is 100: the field distribution places all cataloged output — 1 of 1 reviewed papers — in a single field, 'speech processing / silent-speech interfaces (EMG-to-text)'. Every facet of the Myovox paper, from surface EMG decoding to CTC, Conformer modeling, cross-modal distillation, and LLM reranking, sits inside this one line of work. The sample is tiny, but within it the concentration on silent-speech decoding is total.

Collaboration0/100

The collaboration score is 0, with 0 distinct co-authors recorded and an empty frequent-co-author list; the reviewed Myovox paper lists this researcher as first and corresponding author. The snapshot therefore shows an independently led effort, though a single-paper sample may also reflect incomplete author metadata rather than a confirmed solo work.

Activity—

Ongoing-activity signals (recent publication cadence, per-year output) are not yet available, so this axis is unscored. It would measure whether the researcher is publishing steadily, with 2026 the only year represented in the current catalog.

Reproducibility100/100

The reproducibility score is 100, computed as reproMean 100 across 1 paper; the Myovox review rates its reproducibility readiness as high. The review also calls the methodology rigorous while flagging the narrow single-speaker scope. As with all single-paper scores, this signals strong current practice rather than a settled track record.

Research summary

Varshith Madishetty works in speech processing, focused on silent-speech interfaces that decode speech from surface electromyography (sEMG) signals of the facial muscles. The Evolve catalog currently holds one reviewed paper by this researcher: 'Myovox: Reading Speech from the Muscles of the Face' (arXiv, 2026), on which they are first and corresponding author. The paper reports reducing word error rate on single-subject sEMG speech decoding from 51.17% to 18.53%, which the Evolve review attributes to decoding fixes, a bidirectional Conformer, and LLM reranking; the paper's keyword set also spans cross-modal distillation and CTC. The review judges the work rigorous while noting its narrow single-speaker scope, and rates its reproducibility readiness as high; accordingly, the reproducibility dimension in this profile's radar is scored 100 (reproMean 100 across 1 paper). Field concentration is likewise maximal: the focus score is 100, with the field distribution placing all cataloged output in 'speech processing / silent-speech interfaces (EMG-to-text)'. By contrast, the collaboration score is 0, with zero distinct co-authors recorded and no frequent co-authors listed, so the available picture is of an early, independently led line of work. Output, impact, and activity metrics are not yet computable from the snapshot, so this remains an early-stage profile built from a single reviewed paper; conclusions should be treated accordingly.

Collaboration pattern

The snapshot records 0 distinct co-authors and no frequent co-authors; on the cataloged Myovox paper this researcher is first and corresponding author, so the available evidence points to an independently led effort in EMG-based silent speech. Because the catalog holds only one paper, this pattern may reflect missing author metadata as much as genuine solo work, and no institutional or lab affiliations are on file to contextualize it.

Research keywords

surface electromyographysilent speech interfacesEMG-to-textspeech decodingCTCConformercross-modal distillationLLM rerankingdecoding error correctionsingle-subject decoding

Highlights

51.17% → 18.53% WER on single-subject sEMG speech decoding
The Myovox paper (arXiv, 2026) reports cutting word error rate from 51.17% to 18.53% on single-subject sEMG speech decoding, which the Evolve review attributes to decoding fixes, a bidirectional Conformer, and LLM reranking.
First and corresponding author leading the work
On the cataloged paper, this researcher is listed as first author and corresponding author, indicating an independently led line of work on EMG-based silent speech.
High reproducibility readiness, rigorous but single-speaker
Evolve rates the Myovox paper's reproducibility readiness as high and calls the work rigorous, while noting its narrow single-speaker scope; the profile's reproducibility score is 100 (reproMean 100, 1 paper).

Timeline

2026Published 'Myovox: Reading Speech from the Muscles of the Face' on arXiv as first and corresponding author.
2026Reported reducing single-subject sEMG speech-decoding word error rate from 51.17% to 18.53%.
2026Paper entered the Evolve catalog with a high reproducibility-readiness rating.

Papers in the Evolve catalog

2026-09-17first authorcorresponding
Myovox: Reading Speech from the Muscles of the Face

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 computed from a public bibliographic snapshot (paper source: arXiv); no institutional, ORCID, or country metadata is on file.
The profile is name-resolved from public bibliographic data; independent identity confirmation is not available.
Only 1 reviewed paper underlies this profile, so output, impact, and activity axes are missing and all scores rest on a single sample.
Collaboration data shows 0 distinct co-authors, which may reflect incomplete author metadata rather than confirmed solo authorship.