Audio and speech ML researcher: LALM encoder selection and graph-guided Mamba for audio-visual speech enhancement.
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—
No output score yet: the catalog holds only two 2026 arXiv preprints, so publication volume, venue mix, and trajectory cannot be scored. This axis would measure total and per-year output across indexed venues.
Impact—
No impact score yet: citation counts are not available for this profile, so scholarly uptake cannot be scored. This axis would measure citations, aggregate impact indicators, and how later work references the 2026 preprints.
Focus0/100
The focus score is 0, reflecting an even 50/50 split in the raw field distribution: half of the reviewed work in machine learning for audio and speech processing and half in audio-visual speech enhancement. Both directions sit within speech technology, but neither has yet dominated the two-paper record, so topical concentration is currently minimal rather than deep specialization.
Collaboration41/100
The collaboration score is 41, computed from 7 distinct co-authors across the cataloged record. All seven also appear as frequent collaborators: Hsin-Min Wang, Pei-Jun Liao, Wenze Ren, Kuo-Hsuan Hung, Hung-yi Lee, Guo-Ruei Tseng, and Berlin Chen. On both 2026 papers Lee holds a middle authorship position without corresponding-author status, consistent with contributing inside a team rather than leading the authorship.
Activity—
No activity score yet: with only two 2026 preprints and no multi-year record, publication cadence and consistency cannot be measured. This axis would track output frequency and continuity across recent years.
Reproducibility55/100
Reproducibility scores 55, the mean across 2 Evolve-rated papers, both carrying a medium readiness rating. The G-Mamba paper is described as well-ablated, though small gains lack significance tests and baseline fairness is questioned; the encoder-selection paper shows strong results but is validated on one small backbone and one benchmark. Code or artifact release is not documented in the available data.
Research summary
Hung-Shin Lee works at the intersection of machine learning for audio and speech processing and speech signal processing, with the Evolve catalog documenting two 2026 arXiv preprints split evenly — 50% in each of the two adjacent fields. The first, Correlation-Guided Encoder Selection for Multi-Encoder Large Audio-Language Models, presents a training-free way to choose encoders for multi-encoder large audio-language models from single-encoder score correlations, evaluated on the XARES-LLM benchmark; Evolve's verdict credits strong results while noting validation on one small backbone and one benchmark. The second, G-Mamba: Sparse Graph-Guided Mamba for Audio-Visual Speech Enhancement, fuses graph-guided modeling with Mamba state-space networks for lightweight audio-visual speech enhancement; reviewers praise a thorough ablation but flag small gains without significance tests and questionable baseline fairness. Both papers list Lee as a middle author without corresponding status, and both carry a medium reproducibility readiness rating, averaging 55 across 2 scored papers. The collaboration footprint counts 7 distinct co-authors, including Hsin-Min Wang, Hung-yi Lee, and Berlin Chen — a pattern consistent with team-based research rather than a solo or leading-author profile.
Collaboration pattern
The cataloged record shows a compact, stable collaboration pattern: 7 distinct co-authors in total, all seven of whom also rank as frequent collaborators — Hsin-Min Wang, Pei-Jun Liao, Wenze Ren, Kuo-Hsuan Hung, Hung-yi Lee, Guo-Ruei Tseng, and Berlin Chen. On both 2026 papers Lee contributes as a middle author without corresponding status, consistent with working within an established team rather than leading an independent group.
Research keywords
Highlights
Training-free encoder selection for multi-encoder LALMs
Graph-guided Mamba for lightweight audio-visual speech enhancement
A two-track 2026 record with medium reproducibility readiness
Both 2026 preprints carry a medium reproducibility rating, averaging 55 across the 2 scored papers — reflecting efficient multimodal architectures that reviewers judge partially re-runnable, with caveats around benchmark breadth and statistical rigor.
Timeline
- 2026Published Correlation-Guided Encoder Selection for Multi-Encoder Large Audio-Language Models on arXiv: a training-free encoder-selection method evaluated on the XARES-LLM benchmark, with Lee in a middle authorship position.
- 2026Published G-Mamba: Sparse Graph-Guided Mamba for Audio-Visual Speech Enhancement on arXiv, applying state-space models and graph-guided fusion to lightweight AVSE.
- 2026Entered the Evolve catalog with two medium-reproducibility preprints (mean score 55 across 2 papers), a 7-person distinct-coauthor network, and a 50/50 split across two adjacent speech fields.
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 scores (collaboration, focus, reproducibility) come from a small Evolve bibliometric snapshot covering only the two 2026 arXiv preprints in the catalog.
- Institution, country, and ORCID are not provided; this profile is name-resolved from public bibliographic data and may conflate namesakes.
- Output, impact, and activity axes lack underlying data (no citation counts, no multi-year record) and are left unscored.
- Both papers list a middle authorship with no corresponding role, so claims of research leadership cannot be inferred from this data.