Speech and NLP researcher advancing spoken language models, text-to-speech synthesis, and duplex voice interaction
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
20
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—
Not yet scored: this snapshot lacks the data needed to measure career-wide publication volume, yearly productivity, and first/corresponding authorship share. This axis would aggregate verified output across the researcher's full publication record.
Impact—
Not yet scored: citation-based influence data is insufficient in this snapshot. The axis would measure citation uptake, h-index-relative standing, and real-world adoption of the reviewed speech models; no estimate is offered here.
Focus0/100
Scored 0/100, the lowest concentration: the field distribution splits exactly 50/50 between speech and natural language processing and speech processing (text-to-speech synthesis). Topic counts reinforce the breadth — speech recognition and synthesis (52), NLP techniques (43), speech and audio processing (34), dialogue systems (17), topic modeling (16). Work spans two adjacent speech-AI areas rather than one narrow specialization.
Collaboration68/100
Scored 68/100 from 30 distinct coauthors in the snapshot. Frequent partners include Junyi Ao, Kainan Peng, Mingbo Ma, Shun Zhang, Zhenyu Tang, Xiang Li, Yinghao Li, Yuancheng Wang, Zhizheng Wu, Haizhou Li, Qing He, and Xubo Liu — a network spanning speech synthesis and spoken-language modeling. Both reviewed 2026 papers list middle authorship without corresponding status, consistent with contributor roles inside larger teams.
Activity86/100
Scored 86/100, the strongest dimension. Snapshot per-year counts run 2 (2015), 1 (2017), 3 (2018), 8 (2019), 11 (2020, peak), then 6, 6, 7, 9, 7 through 2025, with 2 logged for 2026. Eighteen works fall within the last three years — a sustained, near-decade stretch of continuous publication without major gaps.
Reproducibility20/100
Scored 20/100 (mean 20 across the 2 papers Evolve has reviewed). Both 2026 arXiv releases carry low reproducibility readiness: the think-aloud spoken-language work depends on proprietary data with no interleaving baseline, and Evolve notes Text-Audiobox compares only against internal baselines. Independent replication of either system is currently difficult.
Research summary
Xutai Ma is a researcher working across speech and natural language processing, with the snapshot's field split exactly even (50/50) between speech-and-NLP and speech processing focused on text-to-speech synthesis. Topic counts place speech recognition and synthesis first (52 works), followed by natural language processing techniques (43), speech and audio processing (34), speech and dialogue systems (17), and topic modeling (16). Activity is the strongest scored dimension (86/100): per-year counts climb from 2 works in 2015 to a 2020 peak of 11, then hold at 6-9 through 2025, with 18 works in the last three years and 2 logged for 2026. Collaboration scores 68/100 across 30 distinct coauthors, with frequent partners including Junyi Ao, Kainan Peng, Zhizheng Wu, and Haizhou Li. The snapshot reports an h-index of 20 and i10-index of 24, though the output and impact dimensions remain unscored for lack of underlying data. Two 2026 arXiv papers sit in the Evolve catalog, both with middle authorship and no corresponding role: Spoken Language Models that Think Aloud, where asynchronous think-aloud decoding cuts reasoning silence from 12.8s to 0.36s at near-chain-of-thought accuracy, and Meta's 3B flow-matching Text-Audiobox for voice dubbing and full-duplex dialogue synthesis. Evolve rates both low on reproducibility readiness (dimension mean 20/100), citing proprietary data and internal-only baselines.
Collaboration pattern
Xutai Ma's collaboration score is 68/100, built on 30 distinct coauthors. Recurring partners include Junyi Ao, Kainan Peng, Mingbo Ma, Shun Zhang, Zhenyu Tang, Xiang Li, Yinghao Li, Yuancheng Wang, Zhizheng Wu, Haizhou Li, Qing He, and Xubo Liu — a roster spanning speech synthesis and spoken-language modeling. Both catalogued 2026 papers list middle authorship without corresponding status, pointing to contributor roles within larger research teams rather than solo-led projects.
Research keywords
Highlights
Async think-aloud slashes spoken-reasoning silence
Alignment-free Text-Audiobox for dubbing and duplex dialogue
A through-line in low-latency, full-duplex voice interaction
Both catalogued 2026 papers attack latency in spoken interaction from different ends: think-aloud decoding removes reasoning silence in spoken language models, while Text-Audiobox targets full-duplex dialogue synthesis — mirroring the snapshot's even 50/50 split between speech-and-NLP and TTS-focused speech processing.
Timeline
- 2015Earliest publications in the snapshot's per-year record: 2 works logged.
- 2019Output steps up to 8 works, beginning a sustained annual cadence.
- 2020Peak year in the snapshot with 11 publications logged.
- 20249 works logged, within the steady 6-9 annual range that holds through 2025.
- 2026Two arXiv preprints enter the Evolve catalog — Spoken Language Models that Think Aloud and Alignment-Free Text-Audiobox — both middle-authored and non-corresponding.
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.
- Identity is resolved from OpenAlex bibliographic data only; institution, country, and ORCID are absent, so namesake merging or splitting is possible.
- Output and impact radar dimensions are unscored because the underlying data is insufficient; the h-index (20) and i10-index (24) appear in the source snapshot and are not independently verified.
- Reproducibility rests on only the 2 papers Evolve has reviewed, both 2026 arXiv preprints rated low reproducibility readiness.
- Per-year publication counts (2015-2026), topic counts, and coauthor statistics come from the same snapshot and have not been audited against primary records.