Veteran graphics and robotics researcher now at peak output in full-duplex speech dialogue and voice-agent AI
4
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
92
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-17) 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 computable from the current snapshot: this axis would measure publication volume, venue quality, and first- or corresponding-author share. No output score is present in the raw data, so it is left unscored rather than estimated.
Impact—
Insufficient data in this snapshot: this axis would measure citation totals, field-normalized citation percentiles, and landmark-paper influence. Although an h-index of 92 and an i10-index of 623 are recorded, no composite impact score was computed, so the axis stays unscored.
Focus0/100
The focus score is 0, the lowest reading on the radar, reflecting an unusually wide topical spread. The OpenAlex field distribution splits 0.5/0.5 between spoken dialogue and voice-agent evaluation and conversational AI, while the topic list spans computer graphics (443 works), 3D shape modeling (295), robot path planning (234), computational geometry (168), and reinforcement learning in robotics (149) — evidence of a career-long pivot from graphics and robotics into speech AI.
Collaboration65/100
Collaboration scores 65, supported by 25 distinct coauthors over the most recent three years. The frequent-coauthor list is anchored by long-time Maryland colleagues Ramani Duraiswami, Sreyan Ghosh, and Puneet Mathur, alongside a newer generation including Vaibhavi Lokegaonkar, Utkarsh Tyagi, and Sonal Kumar. Across the four 2026 catalog papers Manocha holds middle or last author positions, never first, consistent with a senior lab leader sharing the final-author slot with students.
Activity100/100
Activity is the strongest axis at 100. The countsByYear series shows continuous output from 1990 (4 works) to 2026 (42 works to date), with sustained acceleration: 44 works in 2008, 80 in 2020, a record 93 in 2024, and 75 in 2025. Recent3yWorks totals 210, and the Evolve catalog already holds four 2026 arXiv preprints.
Reproducibility55/100
Reproducibility scores 55 (reproMean), computed from 2 papers with reproducibility data in the catalog. Both reviewed speech papers carry a medium repro-readiness rating: SteerDuplex introduces the SteerBench benchmark but is flagged for thin baselines and single-judge scoring, while DuplexSpeechBench-IFEval is judged rigorous in design yet limited to 30 conflicts with no significance testing.
Research summary
Dinesh Manocha, of the University of Maryland, College Park, carries one of the longest continuous publication records in this dataset: an OpenAlex snapshot lists an h-index of 92, an i10-index of 623, and works logged every year from 1990 through 2026. Far from slowing, output has accelerated — countsByYear climbs from 44 works in 2008 to 80 in 2020 and a record 93 in 2024, with 210 works across the most recent three years and 25 distinct coauthors. His topical footprint is broad: OpenAlex topics are led by Computer Graphics and Visualization Techniques (443), 3D Shape Modeling and Analysis (295), Robotic Path Planning Algorithms (234), Computational Geometry and Mesh Generation (168), and Reinforcement Learning in Robotics (149). The Evolve catalog, however, captures his current research front in speech processing, split evenly between conversational AI and spoken-dialogue/voice-agent evaluation. Four 2026 arXiv preprints appear: SteerDuplex (steerable full-duplex dialogue models), DuplexSpeechBench-IFEval (implicit instruction following in voice agents), TEMPO (multi-task post-training for large audio-language models), and VIBE (video instruction-aligned background music generation). Evolve's review of SteerDuplex credits SFT+RL gains of 44.5pp on the new SteerBench while flagging thin baselines and single-judge scoring; DuplexSpeechBench-IFEval is called rigorous but limited by 30 conflicts and no significance tests.
Collaboration pattern
Collaboration is dense and lab-centered: the radar scores it 65, with 25 distinct coauthors in the most recent three years. The frequent-coauthor list is anchored by long-time Maryland colleagues Ramani Duraiswami and Sreyan Ghosh and includes younger researchers such as Puneet Mathur, Vaibhavi Lokegaonkar, Utkarsh Tyagi, and Sonal Kumar. In the 2026 catalog papers Manocha sits in middle or last author slots (never first), consistent with a senior supervisor sharing final-author credit with students.
Research keywords
Highlights
SteerDuplex: making full-duplex speech dialogue steerable with SFT+RL
DuplexSpeechBench-IFEval: a benchmark for implicit instruction following in voice agents
An active 2026 preprint pipeline beyond speech dialogue
Alongside the reviewed speech work, the catalog lists TEMPO (temporally-grounded multi-task post-training for large audio-language models) and VIBE (video instruction-aligned background music generation), both 2026 arXiv preprints with Manocha as last author; neither yet carries an Evolve verdict, keywords, or reproducibility rating.
Timeline
- 1990Earliest year in the countsByYear record, with 4 works — the start of a continuous publishing run that extends through 2026.
- 200844 works recorded in a single year, then the highest annual count in the series and an early sign of the sustained acceleration ahead.
- 2020Annual output reaches 80 works, part of the steep climb through the 2020s that follows years of 60-80 works per year.
- 2024Peak year in the record with 93 works; the surrounding three-year window (2024-2026) totals 210 works, matching the reported recent3yWorks figure.
- 2026Four arXiv preprints in the Evolve catalog — SteerDuplex, DuplexSpeechBench-IFEval, VIBE, and TEMPO — marking the current full-duplex speech and audio-language research front, with 42 works logged year-to-date.
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.
- Bibliometrics (h-index, i10-index, countsByYear, topics) come from an OpenAlex snapshot; identity is name-resolved from public bibliographic data.
- The output and impact radar axes have no computed scores in this snapshot and are left unscored.
- VIBE and TEMPO appear in the catalog without Evolve review verdicts, keywords, or reproducibility ratings.
- Catalog coverage is limited to 2026 arXiv preprints in speech processing, and the 2026 countsByYear figure reflects a partial year.