Kaiqi Fu

—

Unclaimed · AI-structured profileOpenAlex

Speech processing researcher working on streaming ASR, TTS and full-duplex dialogue systems, centered on real-time spoken interaction.

papers here

reviewed by Evolve

first-author

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 output score is available yet: the underlying bibliometric data lacks a validated publication-volume measure for this axis. It would measure annual scholarly output and its consistency across venues and years.

Impact—

Impact is not scored: citation aggregation for this researcher is insufficient in the current data, so the axis is left unscored. It would measure citation-based influence, such as citation counts and the citation trajectory of published work.

Focus0/100

The focus score is 0, reflecting an almost perfectly even three-way split across speech processing (text-to-speech), speech processing (automatic speech recognition), and speech processing / spoken dialogue systems (about 0.33 each). Underlying topic counts — speech recognition and synthesis (19), phonetics and phonology research (14), speech and audio processing (13), speech and dialogue systems (7), EFL/ESL teaching and learning (2) — indicate breadth within speech rather than one narrow specialty.

Collaboration53/100

Collaboration scores 53, computed from 13 distinct coauthors in the recent three-year window. Twelve frequent coauthors appear in the data, and all three catalog papers list Fu as a middle, non-corresponding author, indicating contribution within larger team settings rather than sole or leading authorship. The coauthor roster spans the full-duplex, streaming TTS, and streaming ASR projects, consistent with a group-based research pattern.

Activity52/100

Activity scores 52. Annual counts show a small but sustained record: 1 work in 2019, 4 in 2020, 3 in 2021, 3 in 2022, 4 in 2023, 1 in 2024, and 4 recorded in 2026, with 5 works in the recent three-year window. The 2026 uptick aligns with three arXiv preprints in the Evolve catalog, suggesting renewed posting activity.

Reproducibility58/100

Reproducibility scores 58, a mean computed over 3 papers. Readiness ratings in the catalog are split: the full-duplex floor-taking study is rated high, the X2-NativeCursor text progress tracker medium, and X2Streaming-ASR low. Recurring gaps noted in the reviews include missing significance tests and judge validation, teacher-relative gains measured on synthetic speech, and open cross-system comparability questions.

Research summary

Kaiqi Fu is a speech-processing researcher whose work centers on real-time spoken interaction, spanning streaming automatic speech recognition, text-to-speech, and full-duplex spoken dialogue systems. OpenAlex bibliometric data shows an almost even three-way field distribution across speech processing (text-to-speech), speech processing (automatic speech recognition), and speech processing / spoken dialogue systems (about 0.33 each), with topic mass concentrated in speech recognition and synthesis (19), phonetics and phonology research (14), and speech and audio processing (13). The recorded h-index is 6 with an i10-index of 3, and annual output has been modest but sustained: 1 work in 2019, 4 in 2020, 3 in 2021, 3 in 2022, 4 in 2023, 1 in 2024, and 4 recorded in 2026, with 5 works in the recent three-year window. Three 2026 arXiv papers are in the Evolve catalog: X2Streaming-ASR, a GRPO-trained wait/emit streaming ASR policy cutting commit latency to 24–97 ms; X2-NativeCursor, native-token text progress tracking for streaming codec TTS; and a full-duplex floor-taking study. On all three, Fu is a middle author and not corresponding, consistent with team-based research alongside 13 distinct recent coauthors. Reproducibility signals are mixed, with one catalog paper rated high readiness, one medium, and one low.

Collaboration pattern

Kaiqi Fu's record reflects team-based speech research: 13 distinct coauthors appear in the recent three-year window and 12 frequent coauthors are listed, including Zehan Liu, Rime Wen, Shawn Qin and Hao Wang. All three Evolve-catalog papers are middle-authored and non-corresponding, suggesting a contributor role inside larger project teams spanning full-duplex dialogue, streaming TTS and streaming ASR rather than independent, PI-style work.

Research keywords

streaming ASRstreaming TTSfull-duplex speech modelsturn-takingtext progress trackingcommit latencyreinforcement learning (GRPO)native speech tokensspoken dialogue systemsonline forced alignment

Highlights

Full-duplex speech models take the floor when addressed, not when content demands
An Evolve-reviewed 2026 study argues that full-duplex speech models speak when addressed rather than when content demands; the design is well controlled, but the paper lacks significance tests and judge validation, and its reproducibility readiness is rated high.
Native-token text progress tracking for incremental-text streaming codec TTS
X2-NativeCursor introduces an observer that tracks streaming-TTS text progress directly from native speech tokens, reaching MAE 0.151 versus 1.253 per the Evolve review; the gains are large but teacher-relative and measured on synthetic speech, with medium reproducibility readiness.
Streaming ASR that waits when uncertain and emits when ready
X2Streaming-ASR trains a 3-stage GRPO wait/emit policy for streaming ASR, reaching 24–97 ms commit latency versus 409–585 ms baselines per the Evolve review; the results are strong, though cross-system comparability remains open and reproducibility readiness is rated low.

Timeline

2026Co-authored arXiv study on full-duplex speech models, arguing they take the floor when addressed rather than when content demands; rated high reproducibility readiness despite missing significance tests.
2026Co-authored X2-NativeCursor on arXiv, a native-token observer tracking text progress in incremental-text streaming codec TTS with MAE 0.151 versus 1.253.
2026Co-authored X2Streaming-ASR on arXiv, a 3-stage GRPO-trained wait/emit streaming ASR policy cutting commit latency to 24–97 ms versus 409–585 ms baselines.

Papers in the Evolve catalog

Frequent co-authors

Zehan Liu×2 Rime Wen×2 Shawn Qin×2 Roy Gan×2 Hao Wang×2 Carl Chen×1 Altman Lin×1 Lights Shi×1 Qian Wang×1 Linkai Peng×1 Baorian Nuchged×1 Yuyang Yao×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 OpenAlex data only; no institution, country, or ORCID is verified, so this profile is name-resolved from public bibliographic records and may involve identity merge or split.
Output and impact dimensions lack sufficient underlying data and are left unscored; citation counts are not available in the current snapshot.
The reviewed-paper sample is three 2026 arXiv preprints, all middle-authorship; peer-review status and final venues are not reflected.
The reproducibility score is a mean over only 3 reviewed papers and should be read as indicative rather than comprehensive.