Candice Fan

—

Unclaimed · AI-structured profileOpenAlex

Early-career speech-NLP researcher building full-duplex dialogue agents and audio-emotion/ASR evaluation benchmarks

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

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—

Insufficient data: no total publication count is available in the record, and h-index and i10-index are both listed as 0, so this axis is left unscored. Output would measure cumulative publication volume and its growth across the researcher's career.

Impact—

Insufficient data: the record carries no citation counts, and h-index 0 and i10-index 0 indicate a profile too new for uptake signals. Impact would measure citations received, citing works, and influence on subsequent speech and dialogue research.

Focus100/100

Scores 100/100 for focus. Every available signal points to one narrow area: all three 2026 preprints are speech/audio papers, the field distribution lists a single field (spoken dialogue systems / speech NLP), and topic counts are led by Speech Recognition and Synthesis (3) and Speech and dialogue systems (2), with Emotion and Mood Recognition (1), Speech and Audio Processing (1), and Music and Audio Processing (1) adjacent. No off-topic work appears in the record.

Collaboration48/100

Scores 48/100 for collaboration. The record shows 10 distinct coauthors, and the frequent-collaborator list names Tyler Baumgartner, Brandon Tai, Luc Debaupte, Yi Zhong, Bill Wang, Yunqi Lu, Nikhil Johri, Ruben Aguilar and Lisa Kaelin-Martin. Fan holds the middle author position on all three 2026 papers with no corresponding-author role, suggesting a contributor slot inside larger teams rather than an independent or leading collaboration pattern.

Activity40/100

Scores 40/100 for activity. The per-year counts contain a single entry — 3 works in 2026 — and recent3yWorks is likewise 3, so all visible output is concentrated in one year with no earlier or later activity in the record. This is consistent with an early-stage record whose prior work may not yet be linked, but on the available data the publishing cadence cannot be established beyond 2026.

Reproducibility20/100

Scores 20/100 for reproducibility, computed from 1 repro-scored paper with a mean of 20. The only Evolve-reviewed paper, 'Continue, Adapt, or Yield', carries a low reproducibility-readiness rating, and the review notes that its human-versus-model comparison (68.2% vs 34.8% adaptation on collaborative cues) relies on partly circular labels. The two benchmark papers are not yet scored, so no additional signal exists.

Research summary

Candice Fan is an early-career researcher working on spoken dialogue systems and speech/audio model evaluation, whose public bibliometric record (OpenAlex) shows a compact but thematically concentrated output: three arXiv preprints, all dated 2026, all in the speech and audio space. The most complete picture comes from 'Continue, Adapt, or Yield: In-Turn Adaptation to Overlapping Speech in Full-Duplex Agents', which Evolve reviewed; it introduces Duplex Cue, a 3x3 intent-by-response matrix, and reports that humans adapt on 68.2% of collaborative cues versus 34.8% for PersonaPlex, though Evolve flags partly circular labels. The other two preprints build evaluation assets: VocalAffectBench for vocal emotion recognition in AI audio models, and VoiceCodeBench for exact structured-token recovery in automatic speech recognition. Topic counts confirm the concentration: Speech Recognition and Synthesis (3) and Speech and dialogue systems (2), with Emotion and Mood Recognition (1), Speech and Audio Processing (1), and Music and Audio Processing (1). Across 10 distinct coauthors, Fan sits in the middle author position on all three papers without corresponding-author credit. Headline indicators (h-index 0, i10-index 0) reflect a record too new for citation-based scoring, so the output and impact axes remain unscored in this early-stage profile.

Collaboration pattern

Fan works inside mid-sized teams: 10 distinct coauthors appear in the record, and the frequent-collaborator list includes Tyler Baumgartner, Brandon Tai, Luc Debaupte, Yi Zhong, Bill Wang, Yunqi Lu, Nikhil Johri, Ruben Aguilar and Lisa Kaelin-Martin. On all three 2026 papers Fan is a middle author and never the corresponding author, indicating a steady contributor role within larger groups rather than group leadership.

Research keywords

Full-duplex spoken dialogueTurn-takingOverlapping speechIn-turn adaptationSpeech language modelsConversational AI agentsSpeech emotion recognitionVocal emotion benchmarksAutomatic speech recognition evaluationStructured-token recoveryBenchmark constructionAI audio model evaluation

Highlights

Duplex Cue: a 3x3 intent-by-response test of in-turn adaptation in full-duplex agents
The Evolve-reviewed arXiv 2026 paper 'Continue, Adapt, or Yield' introduces Duplex Cue, a 3x3 intent-by-response matrix, and reports that humans adapt on 68.2% of collaborative cues versus 34.8% for PersonaPlex; Evolve notes partly circular labels and rates reproducibility readiness low.
VocalAffectBench: a benchmark for vocal emotion recognition in AI audio models
An arXiv 2026 preprint that builds an evaluation benchmark for how AI audio models recognize vocal emotion. Evolve has not yet reviewed it, so no verdict, keywords, or reproducibility rating is available for this paper.
VoiceCodeBench: testing exact structured-token recovery in ASR
An arXiv 2026 preprint that proposes a benchmark measuring how exactly automatic speech recognition systems recover structured tokens. It is not yet reviewed by Evolve, so no verdict or reproducibility rating exists.

Timeline

2026Publishes VocalAffectBench (arXiv), a benchmark for vocal emotion recognition in AI audio models.
2026Publishes VoiceCodeBench (arXiv), targeting exact structured-token recovery in automatic speech recognition.
2026Publishes 'Continue, Adapt, or Yield' (arXiv) on in-turn adaptation to overlapping speech in full-duplex agents; Evolve later reviews it, rating reproducibility readiness low.

Papers in the Evolve catalog

2026-09-11
Continue, Adapt, or Yield: In-Turn Adaptation to Overlapping Speech in Full-Duplex Agents

Frequent co-authors

Tyler Baumgartner×3 Brandon Tai×3 Luc Debaupte×3 Yi Zhong×3 Bill Wang×2 Yunqi Lu×1 Nikhil Johri×1 Ruben Aguilar×1 Bill Wang×1 Lisa Kaelin-Martin×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.

Bibliometric scores and topic counts derive from an OpenAlex record only; no institution, country, or ORCID is provided, so this profile is name-resolved from public bibliographic data and may not be unique.
h-index and i10-index are recorded as 0 and no citation counts are available, so output and impact are unscored rather than judged weak.
Only one of the three papers carries an Evolve review; verdicts, keywords, and reproducibility ratings for the two benchmark papers are pending.
All three works are 2026 arXiv preprints with middle authorship; peer-reviewed venue acceptance is unknown from the available data.