Speech and audio ML researcher at CUHK working on agentic audio language models and reproducible quantization analysis
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
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—
No score yet: only two reviewed arXiv papers are linked to this name, so publication volume and author-role distribution cannot be measured. This axis would count indexed outputs per year, venues, and first/corresponding-author share.
Impact—
No score yet: citation counts and downstream uptake are absent from the snapshot. This axis would measure citations to the two catalog papers, reuse of the quantization protocol, and influence on audio-language-model evaluation practice.
Focus0/100
Scores 0/100 for concentration: the field distribution across the two reviewed papers splits exactly 0.5 to speech processing and conversational AI (audio language models) and 0.5 to speech and audio machine learning. With only two papers, this even split reads as breadth across adjacent subfields — one system-building effort, one evaluation and quantization analysis — rather than an established specialty; more indexed papers would be needed before focus can be judged meaningfully.
Collaboration60/100
Scores 60/100: the snapshot records 19 distinct coauthors, with 12 recurring partners — including Lujia Bao, Qian Chen, Chong Deng, Xiangang Li, and Xu Li — a roster size typical of large-team speech-model development rather than small-group work. Both catalog papers carry dense author lists (middle position on the realtime system, last position on the quantization study), indicating an established contributor inside a sizeable collaborative group; small-team or solo patterns are absent from the data.
Activity—
No score yet: both catalog entries are 2026 arXiv preprints and no earlier publication history is linked. This axis would track annual output cadence, preprint-to-venue progression, and sustained engagement across recent years.
Reproducibility60/100
Scores 60/100, averaged over two papers (reproMean 60 across reproPapers 2). The split is stark: the Qwen2-Audio quantization case study is rated high readiness — a careful, honest protocol with explicit separation of text scores, waveform use, and packed memory — while Qwen-Audio-3.1-Realtime is rated low, reporting extensive distillation and GRPO results without ablations, confidence intervals, or independent verification. Reproducibility practice is therefore uneven rather than uniformly weak.
Research summary
Junhao Xu is a researcher at the Chinese University of Hong Kong whose indexed work sits in speech and audio machine learning. The Evolve catalog holds two 2026 arXiv papers that split the field distribution evenly, 0.5 to speech processing and conversational AI (audio language models) and 0.5 to speech and audio machine learning. The first, Qwen-Audio-3.1-Realtime, reports broad agentic voice-interaction gains built on on-policy distillation and GRPO; Evolve notes extensive results but no ablations, confidence intervals, or independent verification, and rates reproducibility readiness low. The second, a Qwen2-Audio post-training quantization case study with Xu as last author, proposes a protocol separating text scores from waveform use and packed-memory effects, evaluated on emotion recognition over RAVDESS; Evolve calls it careful and honest, with high reproducibility readiness. Signals are correspondingly narrow: 19 distinct coauthors with 12 recurring, a collaboration score of 60, and a reproducibility score of 60 averaged over two papers. The even field split yields a focus score of 0, indicating breadth across adjacent subfields rather than concentration. Output, impact, and activity are unscored because citation and volume data are missing, so this profile reads as an early-stage, name-resolved summary from public bibliographic data.
Collaboration pattern
The snapshot links 19 distinct coauthors, 12 of them recurring, including Lujia Bao, Qian Chen, Luyao Cheng, Chong Deng, Yuxiang Kong, Xiangang Li, Xu Li, and Jiaqing Liu. Working as a middle author on the realtime system and last author on the quantization study, Xu operates within a large, stable speech-model team rather than small ad-hoc collaborations; no solo work appears in the catalog.
Research keywords
Highlights
Agentic realtime voice system with distillation and GRPO
Quantization protocol separating text scores from waveform use
Complementary roles: system building and evaluation rigor
Within one year, the catalog shows Xu contributing as a middle author to a large-scale agentic speech system and as last author to a reproducibility-focused quantization analysis. The pairing — one low-readiness system paper, one high-readiness methods paper — frames his visible work as spanning engineering delivery and measurement discipline, though the evidence base is only two papers.
Timeline
- 2026Qwen-Audio-3.1-Realtime appears on arXiv, reporting agentic full-duplex voice interaction via on-policy distillation and GRPO; Xu is a middle-listed author.
- 2026The Qwen2-Audio quantization case study appears on arXiv with Xu as last author, proposing a protocol that separates text scores from waveform use; Evolve rates its reproducibility readiness high.
- 2026Bibliometric snapshot: 19 distinct coauthors with 12 recurring collaborators, and an even 0.5/0.5 split between conversational audio-language-model work and speech/audio machine learning.
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 come from a sparse public bibliographic snapshot: only two reviewed arXiv papers are linked to this name, so scores may understate the full record.
- Junhao Xu is a common name and no ORCID is provided; this profile is name-resolved from public bibliographic data and may merge or split identities.
- Citation counts, h-index, venue history, and career facts are absent from the input; no such numbers are quoted, and output, impact, and activity remain unscored.
- The 0.5/0.5 field distribution is computed from just two 2026 papers, so the focus reading may not generalize beyond this snapshot.