Speech-data and diffusion-LM researcher; first author of YODAS v3, an open 1.1M-hour, 147-language stereo speech corpus
2
papers here
0
reviewed by Evolve
1
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-09-27) 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 score yet: the snapshot lacks publication counts, venue mix and productivity measures. This axis would aggregate total papers, yearly output and authorship positions across the full record.
Impact—
Impact is unscored: citation counts, h-index and per-paper influence are absent from the data. This axis would measure citations, citing works and downstream uptake of the two catalog papers.
Focus—
No focus score: field distribution and topic concentration cannot be computed from the current record, though the two papers span speech data and diffusion language modeling. This axis would measure thematic concentration over time.
Collaboration46/100
The only scored axis: 46/100, computed from 9 distinct coauthors in the bibliometric snapshot. The frequent-coauthor list splits into a speech-processing group — Shinnosuke Takamichi, Sayaka Shiota, Satoru Fukayama, Samuele Cornell, Shinji Watanabe — and Zeyu Michael Li, Bingshuo Qian, Jiayin Liu and Xiang Cheng. The mid-range score reflects a small but specialized network rather than a broad one, spanning both catalog papers.
Activity—
Activity is unscored: the snapshot lacks full publication-year coverage, so recent cadence cannot be computed. This axis would measure publication frequency and research momentum in recent years.
Reproducibility—
The composite reproducibility axis is not yet computed. Per-paper signals exist — Evolve rates both catalog entries 'medium' for repro readiness — but no aggregated score across the full record is available here.
Research summary
William Xingxu Chen is an early-stage researcher whose visible record centers on two fronts: large-scale speech data and diffusion-based language modeling. The Evolve catalog holds two 2026 arXiv papers. The first, YODAS v3, lists him as first author and claims the largest open speech corpus to date — over one million hours (1.1M) of 48 kHz fullband stereophonic audio across 147 languages, assembled through web-scale collection and neural audio codecs; Evolve's review finds the scale and fidelity claims supported while flagging that language tags and transcript quality go unverified. The second, ELF-REG, places him in the middle author list of a method paper that scales continuous diffusion language models to reasoning tasks via autoregressive-teacher representation alignment (REPA) plus regularization, supported by flow matching, knowledge distillation and early-stopping analyses; reviewers credit solid ablations but note scores still trail autoregressive models and that training is single-seed. His collaboration footprint is compact: nine distinct coauthors yield a collaboration score of 46/100, spanning a speech-processing group (Shinnosuke Takamichi, Sayaka Shiota, Satoru Fukayama, Samuele Cornell, Shinji Watanabe) and a second cluster (Zeyu Michael Li, Bingshuo Qian, Jiayin Liu, Xiang Cheng). With no verified institution, country or ORCID, this profile is name-resolved from public bibliographic data.
Collaboration pattern
With nine distinct coauthors, the collaboration footprint is small but two-sided: a speech-processing cluster (Shinnosuke Takamichi, Sayaka Shiota, Satoru Fukayama, Samuele Cornell, Shinji Watanabe) and a second group (Zeyu Michael Li, Bingshuo Qian, Jiayin Liu, Xiang Cheng). The 46/100 collaboration score derives from these nine; both catalog papers are team efforts — first authorship on one, a middle position on the other — with no solo work in the current record.
Research keywords
Highlights
First-authored YODAS v3: 1.1M hours of open, 48 kHz stereo speech across 147 languages
Co-author on ELF-REG: lifting continuous diffusion LMs on reasoning with AR-teacher REPA+REG
A dual-track 2026: web-scale speech data and diffusion-model methodology
Both catalog papers are 2026 arXiv releases rated 'medium' for repro readiness, showing work on the data side (corpus construction at 1.1M-hour scale) and the modeling side (diffusion LM training recipes), with first authorship on the data paper and a middle position on the method paper.
Timeline
- 2026First authors YODAS v3 on arXiv, presenting over one million hours (1.1M) of 48 kHz stereophonic, multilingual speech across 147 languages.
- 2026Co-authors ELF-REG, scaling continuous diffusion language models to reasoning tasks with AR-teacher REPA+REG.
- 2026Both papers enter the Evolve catalog with 'medium' repro-readiness ratings; reviewed strengths include corpus scale and ablation rigor, with verification gaps flagged.
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.
- Name-resolved from public bibliographic data; no ORCID or verified affiliation is available, so identity confidence is moderate for a not-uncommon name.
- Only two papers (both 2026 arXiv) are reviewed in the Evolve catalog; the full publication record may be larger.
- Radar scores come from a limited bibliometric snapshot: only collaboration (46) is scored, and the other five axes lack data.
- Paper-level findings reflect Evolve's review verdicts, not independent verification of the papers' claims.