Junyoung Park

—

Unclaimed · AI-structured profileOpenAlex

Speech technology researcher: Korean lipreading (VSR) and quantized text-to-speech; last author on 2026 arXiv papers

papers here

reviewed by Evolve

first-author

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—

Not yet scored: the profile lacks full publication-count and venue data. This axis would measure total scholarly output, venue quality and consistency of production across years.

Impact—

Not yet scored: citation counts, h-index and venue-level influence are unavailable for this profile. This axis would measure how often and how widely the work is cited and taken up by other researchers.

Focus—

Not yet scored: a complete bibliographic record would be needed to map topic distribution over time. This axis measures how concentrated or diversified the research agenda is across fields and years.

Collaboration32/100

Scored 32/100 from bibliometric data: only 4 distinct coauthors (Se Un Park, Yutae Kim, Hakjun Kim, Taehoon Roh) appear across the cataloged work. Both papers are small-team efforts with Junyoung Park in the last-author position, suggesting a tight, consistent core group rather than a broad network — a pattern typical of an early-stage or specialized research unit.

Activity—

Not yet scored: year-by-year publication records are too sparse to measure cadence. This axis would track publishing frequency and sustained engagement with research over time.

Reproducibility—

Not yet scored: no aggregate artifact or code-availability metric exists yet. It would measure code release, data availability and readiness for independent replication; the two cataloged papers carry medium and high repro-readiness tags respectively.

Research summary

Junyoung Park is a speech technology researcher whose cataloged work centers on making speech models both more personal and more deployable. The Evolve catalog holds two 2026 arXiv papers, each with Park in the last-author position. The first, on the Korean OLKAVS corpus, frames personalized lipreading as visual speech recognition and reports character error rates of 9.95–12.19% against a 26.64% comparison point — a claimed Korean VSR state of the art built on transfer, a speaker census and LoRA-based speaker personalization, with rigorous paired evaluation though prior-method baselines are absent. The second is a cross-architecture study of post-training quantization for text-to-speech, showing that identical bit widths yield different outcomes across architectures, validated on real kernels and blind tests; Evolve rates its reproducibility readiness high while noting proxy-metric caveats. Collaboration data shows a compact footprint: 4 distinct coauthors (Se Un Park, Yutae Kim, Hakjun Kim, Taehoon Roh) recur across both papers. Broader bibliometrics — publication counts, citations, h-index — are not yet available for this profile, and the name is common, so identity is resolved from public bibliographic data only.

Collaboration pattern

Junyoung Park's cataloged output comes from a single small team: 4 distinct coauthors — Se Un Park, Yutae Kim, Hakjun Kim, Taehoon Roh — recur across both 2026 arXiv papers, where Park takes the last-author slot (corresponding status not flagged on either). Reusing the same collaborator set across an AVSR paper and a TTS quantization paper points to a stable core group covering both research lines rather than per-project partnerships.

Research keywords

visual speech recognitionlipreadingOLKAVS corpusspeaker personalizationLoRA parameter-efficient fine-tuningtext-to-speechpost-training quantizationmodel compressionon-device inferencemixed precisionGPTQ

Highlights

Claimed Korean VSR state of the art: 9.95–12.19% CER vs 26.64%
On OLKAVS, the paper reports character error rates of 9.95–12.19% against a 26.64% comparison point, positioning personalized Korean lipreading as visual speech recognition. Evolve's review finds the paired evaluation rigorous but notes the absence of prior-method baselines.
Speaker census and LoRA personalization for Korean lipreading
The same paper combines a speaker census with transfer and LoRA-based parameter-efficient adaptation, tailoring a visual speech recognition model to individual speakers rather than a single generic system.
Same bit width, different outcomes across TTS architectures
The PTQ study applies post-training quantization at identical bit widths to multiple text-to-speech architectures and shows outcomes diverge, using mixed-precision and GPTQ-based pipelines in an on-device inference context.
Real-kernel validation and blind tests earn a high repro-readiness tag
Evolve rates this paper's reproducibility readiness high: the quantization findings are checked on real inference kernels and supported by blind tests, though the review flags reliance on proxy metrics as a caveat.

Timeline

2026Last-author arXiv paper on personalized Korean lipreading on OLKAVS reports 9.95–12.19% CER against 26.64%
2026Last-author arXiv study of post-training quantization for text-to-speech across architectures, validated on real kernels with blind tests
2026Both papers enter the Evolve catalog with medium and high repro-readiness ratings respectively

Papers in the Evolve catalog

Frequent co-authors

Se Un Park×2 Yutae Kim×1 Hakjun Kim×1 Taehoon Roh×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-derived public bibliometric data only; only the collaboration axis had enough data to score — output, impact, focus, activity and reproducibility are unscored.
Profile is name-resolved from public bibliographic data: 'Junyoung Park' is a common Korean name, and no institution, country or ORCID was provided to confirm identity.
Only the two papers reviewed in the Evolve catalog inform this profile; broader publication history is not included.
Both cataloged papers are 2026 arXiv preprints; peer-review status and citation history are not yet reflected.