Speech AI researcher bridging audio-visual speech enhancement architectures and encoder selection for large audio-language models
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-09-19) 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—
Output cannot be scored: this snapshot contains no publication counts or productivity records. The axis would measure total research output and its distribution across venues and years, once fuller bibliometric coverage becomes available for this researcher.
Impact—
Impact cannot be scored: no citation counts, h-index, or influence metrics are present in the data. The axis would measure how widely this researcher's publications are cited and taken up by the field, which cannot yet be estimated.
Focus0/100
The focus score is 0, reflecting an even 50/50 split in the field distribution between "machine learning for audio and speech processing" and "speech signal processing (audio-visual speech enhancement)". In practice this reads as adjacency rather than dispersion: both catalog papers sit on the same speech-AI continuum, one on encoder selection for large audio-language models, the other on graph-guided state-space models for audio-visual speech enhancement.
Collaboration41/100
Collaboration scores 41 out of 100, based on 7 distinct coauthors in the current snapshot. The recurring group includes Hung-Shin Lee, Pei-Jun Liao, Wenze Ren, Kuo-Hsuan Hung, Hung-yi Lee, Guo-Ruei Tseng, and Berlin Chen. On the two catalog papers the researcher takes a last-author position once and a middle position once, neither as corresponding author — a supportive, senior-contributor pattern within a compact team.
Activity—
Activity cannot be scored: the snapshot lacks year-by-year publication volume. The axis would measure recent publishing cadence and continuity, which is currently traceable only through the two 2026 arXiv papers in the reviewed catalog.
Reproducibility55/100
Reproducibility averages 55 across the 2 papers scored, both rated at medium readiness. Evolve's verdicts flag concrete gaps: the encoder-selection work is validated on one small backbone and a single benchmark (XARES-LLM), while the G-Mamba AVSE model is well-ablated but reports small gains without significance tests and raises baseline-fairness questions. Methods appear clearly described, but independent replication would need broader evaluation coverage.
Research summary
Hsin-Min Wang is a speech and audio AI researcher whose reviewed work spans two adjacent areas: machine learning for audio and speech processing, and speech signal processing centered on audio-visual speech enhancement. The Evolve catalog holds two 2026 arXiv papers. The first, CUES, performs correlation-guided encoder selection for multi-encoder large audio-language models without fusion training and was credited with strong results, though validated on one small backbone and a single benchmark. The second, G-Mamba, combines graph-guided fusion, heterogeneous graph attention, and Mamba state-space modeling for lightweight audio-visual speech enhancement; Evolve judged it well-ablated while noting small gains without significance tests and questionable baseline fairness. Bibliometrically, the profile splits evenly across the two fields (0.5 each), giving a focus score of 0; collaboration scores 41 on 7 distinct coauthors including Hung-Shin Lee, Hung-yi Lee, and Berlin Chen; and reproducibility averages 55 across 2 scored papers, both at medium readiness. Authorship positions — last on one paper, middle on the other, neither corresponding — suggest a senior supporting contributor within a compact team in this snapshot. Output, impact, and activity remain unscored for lack of data.
Collaboration pattern
With 7 distinct coauthors and a collaboration score of 41, the pattern is a compact, recurring team rather than a wide network: Hung-Shin Lee, Pei-Jun Liao, Wenze Ren, Kuo-Hsuan Hung, Hung-yi Lee, Guo-Ruei Tseng, and Berlin Chen recur across the snapshot. On the two catalog papers the researcher appears in a last-author role once and a middle position once, neither as corresponding author, consistent with a senior supporting contributor.
Research keywords
Highlights
Training-free encoder selection for multi-encoder large audio-language models
Sparse graph-guided Mamba for lightweight audio-visual speech enhancement
Timeline
- 2026Enters the Evolve catalog with CUES, a training-free, correlation-guided encoder-selection method for multi-encoder large audio-language models (last author, arXiv).
- 2026Co-authors G-Mamba, applying sparse graph-guided Mamba modeling to lightweight audio-visual speech enhancement (middle author, arXiv).
- 2026Both catalog papers land in 2026 across two adjacent speech fields, each carrying weight 0.5 in the profile's field distribution.
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 dimensions are computed from OpenAlex-based public bibliometric data; citation counts, output volume, and yearly activity were not available, so output, impact, and activity are unscored.
- No institution, country, or ORCID was provided; the profile is name-resolved from public bibliographic data and identity confirmation remains pending.
- Coverage is limited to two 2026 arXiv preprints reviewed by Evolve; earlier or peer-reviewed work outside this catalog is out of scope.
- Reproducibility (mean 55) is averaged over only 2 scored papers, so the estimate is coarse.