Berlin Chen

—

Unclaimed · AI-structured profileOpenAlex

Speech researcher and last author on low-resource Taiwanese Hokkien ASR and graph-guided Mamba speech enhancement

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

—

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-09-20) 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 is not scored: the current snapshot lacks verified publication counts, venue metadata, and total output volume beyond the two Evolve-reviewed 2026 preprints. This axis would measure annual publication rate and overall corpus size once indexed records are complete.

Impact—

Impact is not scored: no citation counts, h-index, or downstream uptake data exist in this snapshot. This axis would measure the citation influence and adoption of the two cataloged papers across the speech research community.

Focus0/100

The radar scores focus at 0, and the raw field distribution explains why: the cataloged output splits exactly 50/50 between speech processing/automatic speech recognition and speech signal processing/audio-visual speech enhancement. Rather than one concentrated line, the two papers span two adjacent speech subfields — low-resource ASR over frozen Whisper, and lightweight audio-visual enhancement — a deliberately broad, speech-centric portfolio rather than scattered drift.

Collaboration46/100

Collaboration scores 46 out of 100. The raw metric behind it is 9 distinct co-authors across the two cataloged papers, on both of which Berlin Chen sits in the last (senior) author position. Recurring names include Hung-Shin Lee, Hsin-Min Wang, Tien-Hong Lo, Yu-Sheng Tsao, Hung-Yang Sung, Chien-Chun Wang, Guo-Ruei Tseng, and Yung-Chang Hsu — a compact, stable team rather than a wide external network.

Activity—

Activity is not scored: the snapshot contains no full publication history or year-by-year records beyond the two 2026 papers. This axis would measure research cadence, submission rhythm, and sustained productivity across recent years.

Reproducibility55/100

Reproducibility scores 55, the mean over 2 rated papers, each tagged medium repro-readiness. Evolve describes both studies as well-ablated — T-SANDHI reports consistent relative CER gains of up to 12.27% on a frozen Whisper — yet both reviews withhold full confidence over missing significance testing and questionable baseline fairness. Mid-range readiness matches strong ablations that still lack statistical rigor.

Research summary

Berlin Chen is a speech technology researcher whose Evolve-cataloged output centers on two 2026 arXiv preprints, in both of which he holds the last (senior) author position. The first, T-SANDHI, targets low-resource automatic speech recognition for Taiwanese Hokkien, proposing a tone sandhi-aware adaptive network with decoupled hybrid injection over a frozen Whisper model; Evolve's review reports consistent relative character-error-rate gains of up to 12.27% while flagging significance testing, baseline fairness, and novelty claims for scrutiny. The second, G-Mamba, applies sparse graph-guided Mamba state-space modeling with heterogeneous graph attention to lightweight audio-visual speech enhancement, judged well-ablated although its small gains lack significance tests. The snapshot splits the field profile evenly — 0.5/0.5 between speech processing/ASR and speech signal processing/AVSE — and the radar scores collaboration at 46 across 9 distinct co-authors and reproducibility at 55 averaged over 2 rated papers, both tagged medium repro-readiness. Recurring co-authors include Hung-Shin Lee, Hsin-Min Wang, Tien-Hong Lo, and Yu-Sheng Tsao. Output, impact, and activity lack sufficient data and are unscored. With no institution, country, or ORCID on file, this profile is name-resolved from public bibliographic data.

Collaboration pattern

Berlin Chen works within a compact team: 9 distinct co-authors span the two 2026 papers, on both of which he is last author. The recurring group — Hung-Shin Lee, Hsin-Min Wang, Tien-Hong Lo, Yu-Sheng Tsao, Hung-Yang Sung, Chien-Chun Wang, Guo-Ruei Tseng, Yung-Chang Hsu — covers both speech recognition and speech enhancement, consistent with a stable lab-style cluster rather than dispersed external collaboration. Radar collaboration score: 46/100.

Research keywords

automatic speech recognitionlow-resource languagesTaiwanese Hokkientone sandhiparameter-efficient fine-tuningWhisper (speech foundation models)audio-visual speech enhancementMamba / state space modelsgraph neural networksheterogeneous graph attentionmultimodal fusionlightweight models

Highlights

T-SANDHI: tone sandhi-aware ASR for low-resource Taiwanese Hokkien
As last author, contributed to a dual-stream gated injection design over a frozen Whisper model that decouples tone sandhi from rare citation tones; Evolve's review finds consistent relative CER gains of up to 12.27%, well-ablated though short on significance testing and baseline-fairness checks.
G-Mamba: sparse graph-guided Mamba for lightweight audio-visual speech enhancement
As last author, contributed to combining graph-guided fusion with Mamba state-space modeling and heterogeneous graph attention to build a lightweight AVSE model; Evolve judges it well-ablated, while noting the gains are small, lack significance tests, and raise baseline-fairness questions.

Timeline

2026T-SANDHI, a tone sandhi-aware adaptive network with decoupled hybrid injection over frozen Whisper for low-resource Taiwanese Hokkien ASR, appeared on arXiv with Berlin Chen as last author.
2026G-Mamba, a sparse graph-guided Mamba model for lightweight audio-visual speech enhancement, appeared on arXiv, again with Berlin Chen in the last author position.
2026Both papers entered the Evolve catalog with medium repro-readiness; reproducibility averaged 55 across the two, with 9 distinct co-authors on record.

Papers in the Evolve catalog

Frequent co-authors

Hung-Yang Sung×1 Chien-Chun Wang×1 Tien-Hong Lo×1 Yu-Sheng Tsao×1 Yung-Chang Hsu×1 Berlin Chen×1 Guo-Ruei Tseng×1 Hung-Shin Lee×1 Hsin-Min Wang×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.

Bibliometrics come from OpenAlex-backed radar scores; only the two Evolve-reviewed 2026 arXiv preprints are in scope, so venue, peer-review status, and citation data are absent.
No institution, country, or ORCID is available; 'Berlin Chen' is a name shared by multiple speech researchers, so this profile is name-resolved from public bibliographic data and may carry identity-conflation risk.
Output, impact, and activity axes are unscored because the provided data contains no publication counts, citation metrics, or year-by-year records.
Both cataloged papers are arXiv preprints with no confirmed peer-review venue; evolveVerdicts reflect editorial review, not accepted-paper status.