NVIDIA speech AI researcher focused on full-duplex speech-to-speech models, tool calling, and spoken dialogue systems.
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
17
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-15) 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: this snapshot lacks a verified total publication count, so yearly volume (e.g., 14 works in 2024, 15 in 2025) cannot yet be normalized into a scored output axis. The axis would measure production volume and consistency over the full career.
Impact—
Impact is unscored: h-index (17) and i10-index (31) are recorded, but aggregate citation counts and per-paper citations are absent from this snapshot, so citation influence cannot yet be computed for this axis.
Focus0/100
Focus scores 0/100, the lowest reading on this radar, even though the field distribution looks concentrated: 0.5 weight in speech processing and conversational AI and 0.5 in spoken dialogue systems (cs.CL), with topic counts led by speech recognition and synthesis (76) and speech and audio processing (59). The mismatch with the concentrated field labels suggests the axis measures something beyond field tags — possibly venue or topic spread — so this score should be read cautiously.
Collaboration83/100
Collaboration scores 83/100 on the strength of 64 distinct co-authors within a three-year window that produced 39 works — a strongly team-based industry pattern. Recurring collaborators include Edresson Casanova, Jagadeesh Balam, Elena Rastorgueva, Nikhil Srihari, Oluwatobi Olabiyi, Slyne Deng, Travis Bartley, Sanjay Chauhan, Chen Chen, and Viet Anh Trinh, consistent with large-group speech-model engineering at NVIDIA rather than solo authorship.
Activity100/100
Activity is scored 100/100: 39 works in the last three years, with per-year counts accelerating from 3-6 works annually (2014-2023) to 14 in 2024, 15 in 2025, and 10 already recorded in 2026. Those recent three years account for roughly half of the 78 works recorded since 2014, marking the most productive stretch of this profile.
Reproducibility38/100
Reproducibility scores 38/100, derived from Evolve's review of two papers (reproMean 38). The open NemotronLabs VoiceChat full-duplex model rates medium readiness, while the frontend-backend tool-calling architecture rates low: its benchmark gains come without significance testing. Both are arXiv preprints, so code, data, and evaluation pipelines are not yet independently validated.
Research summary
Zhehuai Chen is a speech AI researcher at NVIDIA (United States) whose OpenAlex-resolved profile centers on speech processing and conversational AI, split evenly (0.5 each) between speech processing / conversational AI and spoken dialogue systems (cs.CL). Topic counts are led by speech recognition and synthesis (76), speech and audio processing (59), and speech and dialogue systems (31), with smaller clusters in topic modeling (17) and NLP techniques (15). The snapshot records an h-index of 17 and an i10-index of 31, but no aggregate citation totals, so the output and impact axes remain unscored. Activity is the strongest signal: 39 works in the last three years, accelerating from 3-6 per year (2014-2023) to 14 in 2024, 15 in 2025, and 10 in 2026 — together roughly half of the 78 works recorded since 2014. Collaboration scores 83/100, reflecting 64 distinct co-authors and a recurring NVIDIA core such as Edresson Casanova, Jagadeesh Balam, and Elena Rastorgueva. Evolve has reviewed two 2026 arXiv papers, both on full-duplex speech-to-speech models with tool calling: an open model with native tool calling but weak argument accuracy (42.2%, Pass@1 33.0%), and a frontend-backend delegation architecture with strong benchmarks but no significance testing. Mean reproducibility readiness across the two is 38/100.
Collaboration pattern
Collaboration is intense and team-based: 64 distinct co-authors across 39 works in the last three years (score 83/100). The recurring core — Edresson Casanova, Jagadeesh Balam, Elena Rastorgueva, Nikhil Srihari, Oluwatobi Olabiyi, Slyne Deng, Travis Bartley, Sanjay Chauhan, Chen Chen, and Viet Anh Trinh — points to a stable NVIDIA speech group producing system papers collectively rather than solo. The co-author list also contains the researcher's own name, a likely name-disambiguation artifact.
Research keywords
Highlights
Open full-duplex speech-to-speech model with native tool calling
Frontend-backend delegation architecture for duplex speech tool calls
Sustained output acceleration since 2024
Annual work counts in the OpenAlex snapshot rise from 3-6 per year (2014-2023) to 14 in 2024 and 15 in 2025, with 10 already recorded in 2026 — 39 works over the last three years, roughly half of the 78 recorded since 2014.
Timeline
- 2014First recorded work appears in the OpenAlex snapshot (1 publication).
- 2018Output reaches 6 works, then settles into a steady 3-6 per year through 2023.
- 2024Publication count jumps to 14 works, beginning the most productive phase of the profile.
- 2025Career-high 15 works recorded in the snapshot.
- 202610 works recorded so far; two full-duplex speech-to-speech papers with tool calling enter the Evolve catalog.
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 metrics come from an OpenAlex snapshot only; aggregate citation totals are absent, leaving output and impact unscored.
- Identity is name-resolved from public bibliographic data; confirm via ORCID 0000-0003-4400-5340.
- Affiliation metadata is internally inconsistent (Nvidia, United States, with country code GB) and should be treated as unverified.
- Highlights and the reproducibility score rest on only two Evolve-reviewed 2026 arXiv preprints; earlier work is not reviewed in-catalog.