Lei Xie

Northwestern Polytechnical University · China

Unclaimed · AI-structured profileOpenAlex

Speech synthesis and clinical audio-language modeling researcher at Northwestern Polytechnical University

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

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 is available yet: the record lists only 6 recent three-year works, without a verified full publication count, so cumulative productivity cannot be measured. This axis would total peer-reviewed and preprint output over the career.

Impact—

Impact cannot be scored: the record reports h-index and i10-index of 0 with no citation counts by year, reflecting missing citation linkage for these 2026 preprints rather than measured influence. This axis would track citations, h-index growth, and citing works.

Focus0/100

The focus score is 0, reflecting an evenly split profile: the field distribution assigns 0.333 each to speech synthesis / speech technology, speech processing / clinical NLP (audio-language modeling), and speech synthesis (text-to-speech, machine learning). OpenAlex topics reinforce the breadth — Speech Recognition and Synthesis (5), Speech and Audio Processing (4), Speech and dialogue systems (2), NLP Techniques (1), Phonetics and Phonology (1) — indicating two parallel research lines rather than one dominant specialty.

Collaboration53/100

Scored 53, the collaboration axis reflects a moderate, team-centered pattern: 13 distinct coauthors appear in the recent three-year window, with recurring partners including Ziyu Zhang, Hanzhao Li, and Tianlun Zuo. On all three catalog papers Lei Xie appears as last author without a corresponding-author flag, consistent with a senior supervisory role in a group producing both speech synthesis systems and clinical audio-language studies.

Activity57/100

Scored 57, activity is recent and concentrated: OpenAlex records 6 works in the last three years, with the yearly counts attributed entirely to 2026, and all three Evolve-reviewed papers are 2026 arXiv preprints. This suggests a burst of near-simultaneous output across two lines — controllable TTS and clinical audio understanding — though the observation window is short and earlier-year counts are absent.

Reproducibility20/100

Scored 20, the lowest present axis: reproducibility was assessed on 3 papers with a mean readiness of 20, and all three are rated low. Evolve's verdicts cite untuned baselines, missing significance tests, and ablations where single modules match the full combination — so while the reported results are judged plausible, independent reimplementation risk is high across both the TTS and clinical pipelines.

Research summary

Lei Xie is a researcher at Northwestern Polytechnical University, China, whose work spans speech synthesis and clinical audio-language modeling. The OpenAlex record lists 6 works in the most recent three-year window, all dated 2026, with 13 distinct coauthors. Topic counts concentrate on Speech Recognition and Synthesis (5) and Speech and Audio Processing (4), with smaller counts in speech and dialogue systems (2), NLP techniques (1), and phonetics and phonology (1). Three 2026 arXiv preprints are reviewed in the Evolve catalog, on each of which Lei Xie is the last author — a senior-position pattern consistent with group leadership. One line of work is controllable text-to-speech: an NVV-aware DiTAR system reported first place overall (62.786) in ISCSLP 2026 NVV Track 2, and DiTAR+ targets more robust zero-shot autoregressive diffusion TTS through dilated context expansion and acoustic masking. The other line is clinical speech understanding: a comparative study of end-to-end versus cascaded ASR+LLM pipelines for long-form clinical SOAP generation reporting 0.52 versus 0.29 Concept F1. Evolve rates reproducibility low across all three papers (mean 20), citing untuned baselines, missing significance tests, and ablations where single modules match the full system. Citation-impact metrics are not yet linked to this record.

Collaboration pattern

Lei Xie works in a group setting: 13 distinct coauthors appear in the recent three-year window, with recurring partners including Ziyu Zhang, Hanzhao Li, Tianlun Zuo, Yun Chen, and Taihui Wang. On all three Evolve-reviewed papers Lei Xie is the last author without the corresponding-author flag, consistent with a senior supervisory role. The coauthor pool spans both research lines — controllable TTS systems and clinical audio-language modeling — suggesting one team serving two application tracks.

Research keywords

zero-shot text-to-speechautoregressive diffusion transformer (DiTAR)nonverbal vocalizationscontrollable speech generationclinical SOAP note generationend-to-end speech understandinglong-form audioaudio-language modelsinstruction tuningGRPO / RLAIFclassifier-free guidancespeaker similarity

Highlights

First overall in ISCSLP 2026 NVV Track 2 with NVV-aware DiTAR
A nonverbal-vocalization-aware DiTAR pipeline combining long-tail data augmentation, best-of-N sampling, and classifier-free guidance claimed first place overall (62.786) in ISCSLP 2026 NVV Track 2. Evolve rates the engineering credible with honest ablations, though the gain attributable to the architecture itself was not isolated.
DiTAR+ targets robust zero-shot autoregressive diffusion TTS
DiTAR+ introduces dilated context expansion and acoustic attention masking to make zero-shot autoregressive diffusion speech synthesis more robust, with reported gains in speaker similarity. Evolve judges the gains to look real, but notes the paper's own ablation shows single modules matching the full combination.
End-to-end audio-to-SOAP beats cascaded ASR+LLM in comparative study
On long-form clinical audio, the end-to-end audio-language approach reportedly reaches 0.52 Concept F1 versus 0.29 for the cascaded ASR+LLM baseline, supported by instruction tuning and GRPO/RLAIF. Evolve finds the claim plausible but flags untuned baselines, no significance tests, and an unevaluated OPD stage.

Timeline

2026Released DiTAR+, improving robustness of zero-shot autoregressive diffusion TTS via dilated context expansion and acoustic masking (last author)
2026Entered ISCSLP 2026 NVV Challenge Track 2 with an NVV-aware DiTAR system, claiming first place overall at 62.786 (last author)
2026Published a comparative study of end-to-end versus cascaded ASR+LLM pipelines for clinical SOAP generation, reporting 0.52 versus 0.29 Concept F1 (last author)
2026OpenAlex records 6 works in the recent three-year window, all dated 2026, with 13 distinct coauthors

Papers in the Evolve catalog

Frequent co-authors

Ziyu Zhang×2 Hanzhao Li×2 Tianlun Zuo×2 Yun Chen×1 Taihui Wang×1 Qicong Xie×1 Rilin Chen×1 Zhixian Zhao×1 Ziyu Zhang×1 Mingchen Shao×1 Wenjie Tian×1 Longhao Li×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.

Bibliometrics are computed from an OpenAlex record only; Lei Xie is a common name, so this profile is name-resolved from public bibliographic data without ORCID verification.
h-index and i10-index are reported as 0 with no per-year citation counts, so the output and impact dimensions are unscored rather than evaluated as zero.
All three catalog papers are 2026 arXiv preprints; no peer-reviewed venue records appear in this snapshot.
The reproducibility assessment covers only the 3 Evolve-reviewed papers.