UCLA speech researcher advancing parameter-efficient adaptation and hallucination mitigation for ASR foundation models
2
papers here
2
reviewed by Evolve
1
first-author
0
corresponding
—
works · OpenAlex
7
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-09) 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 yet scored: the snapshot lacks a per-year publication-volume metric beyond the OpenAlex list, which attributes 32 works to this identity in the last three years. This axis would measure publishing volume and consistency over time.
Impact—
Impact is not yet scored because citation-influence data is insufficient in this snapshot; the profile lists h-index 7 and i10-index 5 without field-normalized benchmarks. This axis would measure citation reach and scholarly influence relative to peers.
Focus0/100
Focus scores 0, reflecting a sharply split topic profile: speech topics total 66 mentions (Speech Recognition and Synthesis 31, Speech and Audio Processing 26, Speech and Dialogue Systems 9), while a second cluster covers Logic, Programming, and Type Systems (18) and Formal Methods in Verification (14). Within the reviewed work, field distribution splits 0.5/0.5 between ASR and ML-for-ASR.
Collaboration51/100
Collaboration scores 51, a mid-range value. The OpenAlex identity records 12 distinct coauthors against 32 works over the last three years — about one new collaborator per 2.7 works — indicating a compact, stable network rather than a broad one. Recurring partners include Mohan Shi, Zilai Wang, Zihan Wang, Kaiyuan Zhang, Eray Eren, and Abeer Alwan, who anchors the speech-recognition line at UCLA.
Activity100/100
Activity scores 100, the maximum. OpenAlex countsByYear shows a steep ramp — 3 works in 2023, 9 in 2024, 11 in 2025, and 12 in 2026 — totaling 32 works in the last three years. Both Evolve-reviewed papers are 2026 arXiv preprints, one first-author, confirming an intense, ongoing publication cadence concentrated entirely in the most recent window.
Reproducibility78/100
Reproducibility scores 78 (reproMean) across the 2 papers Evolve has reviewed. The first-author AURA paper is rated high readiness, reporting an 89.2%→1.9% hallucination reduction with roughly 500x fewer trainable parameters than LoRA; the co-authored Encoder Awakening paper is rated medium, with the verdict citing modest MyST gains and thin statistical reporting.
Research summary
Natarajan Balaji Shankar is a researcher at the University of California, Los Angeles working on automatic speech recognition (ASR), speech foundation models, and parameter-efficient adaptation. The OpenAlex-based profile records an h-index of 7 and i10-index of 5, with activity concentrated in 2023–2026: 3, 9, 11, and 12 works per year respectively, 32 works in the last three years, and 12 distinct coauthors. Topic data splits between speech (Speech Recognition and Synthesis, 31; Speech and Audio Processing, 26; Speech and Dialogue Systems, 9) and a formal-methods cluster (Logic, Programming, and Type Systems, 18; Formal Methods in Verification, 14), yielding a focus score of 0. Evolve has reviewed two 2026 arXiv preprints. The first-author AURA paper introduces uncertainty-routed activation editing that cuts Whisper non-speech hallucination from 89.2% to 1.9% using roughly 500x fewer parameters than LoRA, and is rated high for reproducibility readiness. A co-authored paper, Encoder Awakening via Adapters, proposes two-stage encoder-adapter awakening followed by joint fine-tuning for domain-shifted ASR with Speech-LLMs, rated medium for reproducibility. Output and impact dimensions remain unscored pending sufficient citation data. Collaboration sits mid-range at 51, with a compact recurring partner set including Mohan Shi, Zilai Wang, Zihan Wang, Kaiyuan Zhang, Eray Eren, and Abeer Alwan.
Collaboration pattern
Collaboration scores 51 (mid-range) with 12 distinct coauthors over 32 recent-3y works. The recurring partner set is compact and stable: Mohan Shi, Zilai Wang, Zihan Wang, Kaiyuan Zhang, and Eray Eren appear repeatedly, with Abeer Alwan anchoring the UCLA speech line. Both catalogued 2026 papers are team efforts, and the first-author AURA work draws on the same core group, suggesting a lab-centered rather than broadly distributed collaboration pattern.
Research keywords
Highlights
AURA: uncertainty-routed activation editing slashes Whisper hallucinations
Encoder Awakening: adapter-based domain adaptation for Speech-LLMs
Timeline
- 2023Activity begins to ramp: OpenAlex countsByYear records 3 works in 2023, the start of a sustained three-year climb toward 32 recent works.
- 2024Output accelerates to 9 works, with the topic profile consolidating around speech recognition, audio processing, and dialogue systems.
- 202511 works recorded; the recent three-year coauthor network reaches 12 distinct collaborators.
- 2026Two arXiv preprints enter the Evolve catalog — first-author AURA (high reproducibility readiness) and co-authored Encoder Awakening via Adapters (medium) — alongside 12 works counted for the year.
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 and counts are computed from OpenAlex only (identity: openalex); citation-impact data is insufficient, so output and impact dimensions are unscored.
- The OpenAlex countsByYear series spans 1986–2026 despite an h-index of 7; pre-2018 entries likely reflect name ambiguity or merged records and should be treated cautiously.
- Only 2 papers have been Evolve-reviewed; the reproducibility score (78) rests on this small sample.
- The frequent-coauthor list includes the researcher's own name, a deduplication artifact in the source data.