Speech-processing researcher on Alibaba's Qwen-Audio line: LLM-based ASR and agentic full-duplex voice interaction
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-16) 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 publication-count or venue data is available yet, so this axis is unscored. Once populated, it would measure the volume, consistency, and venue spread of this researcher's publications over time.
Impact—
No citation counts, h-index, or uptake statistics are in the current snapshot, so impact is not scored. This axis would measure how widely the reviewed papers are cited and built upon by later work.
Focus0/100
The focus axis scores 0. The underlying fieldDist metric shows an even 0.5/0.5 split between two labels — speech processing (automatic speech recognition) and speech processing and conversational AI (audio language models) — so no single field label dominates. In practice the two buckets are adjacent: one paper is an ASR system report, the other an agentic audio-language interaction system.
Collaboration83/100
Collaboration is the strongest measured axis at 83/100. The bibliometric raw data records 62 distinct coauthors, and 12 recurring partners are listed in the profile inputs — including Xiangang Li, Zhifu Gao, Chuanmeng Bian, Bo Gong, and Qing Gu — consistent with the large-team authorship pattern of industrial LLM speech reports. Both cataloged papers sit in middle author positions rather than sole or corresponding roles.
Activity—
Per-year publication rhythm is absent from this snapshot, so activity is unscored. It would measure recent output cadence — papers per year — and whether that pace is accelerating or slowing.
Reproducibility20/100
The reproducibility axis scores 20, computed as reproMean 20 across 2 papers (reproPapers 2). Both reviewed papers carry a low repro-readiness rating: per Evolve, the Qwen-Audio-3.0-ASR report leans on largely internal test sets, and Qwen-Audio-3.1-Realtime omits ablations, confidence intervals, and independent verification, so claims should be treated as unverified until third-party checks appear.
Research summary
Xu Li is a speech-processing researcher whose Evolve catalog footprint centers on the Qwen-Audio model family, spanning automatic speech recognition and agentic audio-language systems. The bibliometric snapshot splits this work evenly (0.5/0.5) between two adjacent field labels — speech processing (automatic speech recognition) and speech processing and conversational AI (audio language models) — which also explains the focus axis score of 0, since no single field dominates. Both cataloged papers are 2026 arXiv releases with middle-position authorship. The Qwen-Audio-3.0-ASR Technical Report, described by Evolve as Alibaba's MoE LLM-ASR effort, claims state-of-the-art Chinese and English recognition plus strong multilingual results, but mostly on internal test sets that cannot be independently verified. Qwen-Audio-3.1-Realtime reports broad agentic voice gains via on-policy distillation and GRPO, yet omits ablations, confidence intervals, and independent verification. Both papers are rated low on reproducibility readiness, and the reproducibility dimension lands at 20 (reproMean 20 over 2 papers). Collaboration is the strongest measured axis at 83, resting on 62 distinct coauthors, with recurring partners including Xiangang Li, Zhifu Gao, and Chuanmeng Bian. Because Xu Li is a common name and no institution, country, or ORCID is supplied, this profile is name-resolved from public bibliographic data and should be treated as an early-stage record.
Collaboration pattern
The collaboration axis (83/100) reflects a wide industrial-lab network: 62 distinct coauthors in the raw bibliometrics, with 12 frequent collaborators listed, including Xiangang Li, Zhifu Gao, Chuanmeng Bian, Bo Gong, and Qing Gu. Both cataloged papers place Xu Li in middle author positions on large-team technical reports rather than first or corresponding author, a pattern typical of big-model speech teams at Alibaba's Qwen-Audio effort.
Research keywords
Highlights
LLM-based MoE ASR: Qwen-Audio-3.0 technical report
Agentic full-duplex voice: Qwen-Audio-3.1-Realtime
Reproducibility is the open question across the catalog
Both reviewed papers carry low repro-readiness; the reproducibility dimension scores 20 (reproMean 20 over 2 papers). Evolve flags missing ablations and internal-only evaluation as the main gaps.
Timeline
- 2026Qwen-Audio-3.0-ASR Technical Report appears on arXiv; Evolve's verdict notes claimed SOTA Chinese/English and multilingual results, mostly on unverifiable internal test sets.
- 2026Qwen-Audio-3.1-Realtime follows, pushing the Qwen-Audio line toward agentic full-duplex voice interaction via on-policy distillation and GRPO.
- 2026Evolve reproducibility review rates both papers low; the reproducibility dimension stands at 20 (reproMean over 2 papers).
- 2026Bibliometric snapshot records 62 distinct coauthors and a collaboration score of 83, with 12 frequent collaborators listed.
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 come from public bibliographic data (OpenAlex-style) only; Xu Li is a common name and no ORCID, institution, or country is provided, so the profile is name-resolved and may merge or split identities.
- Output, impact, and activity dimensions are unscored (dimMissing): this snapshot contains no publication counts, citation metrics, or per-year activity data.
- Only 2 papers, both 2026 arXiv technical reports, are in the reviewed Evolve catalog; journal or conference records are likely under-sampled.
- The reproducibility score (reproMean 20 over 2 papers) reflects Evolve's verdicts on these papers only, not the researcher's full body of work.