Early-career speech AI researcher: streaming transcription and tool calls for full-duplex speech-to-speech models
2
papers here
1
reviewed by Evolve
2
first-author
0
corresponding
—
works · OpenAlex
1
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-09-27) 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 scorable: no aggregated output metric is available beyond 4 works in the last three years. This axis would measure total publication volume, venue mix, and the output trend over the research career.
Impact—
Not yet scorable: the only impact signals available are h-index 1 and i10-index 0. This axis would measure citations, downstream adoption, and influence on later full-duplex speech research.
Focus100/100
A maximal 100 reflects an exceptionally narrow record: OpenAlex classifies the 1 tagged publication under a single field, 'speech processing and conversational AI', and the topic profile stays tightly clustered — Speech Recognition and Synthesis (4), speech and dialogue systems (3), Speech and Audio Processing (2), with Topic Modeling (1) and Phonetics and Phonology Research (1) as the only edge topics.
Collaboration62/100
Scores 62. The record shows 21 distinct coauthors, and 12 names recur across the work — among them Edresson Casanova, Piotr Zelasko, and Boris Ginsburg — indicating contribution inside a large, stable speech team rather than isolated projects. Both catalog papers are first-authored but non-corresponding, consistent with a hands-on contributor within a group effort on full-duplex speech models.
Activity47/100
Scores 47, matching a short, recent record: 4 works in the last three years, split 1 in 2025 and 3 in 2026 — an accelerating but still small cadence. Both 2026 catalog entries are first-author arXiv preprints, indicating an active current phase whose volume has not yet accumulated.
Reproducibility55/100
Averages 55 across the single reviewed paper with data. Evolve rates the streaming-transcription work 'medium': it reports a concrete 10.21% WER from a parallel ASR head with turn-taking preserved, but the verdict cites 'thin statistics', so the evidence rests on limited evaluation rather than broad, well-powered benchmarking.
Research summary
Ke Hu (profile name-resolved from public OpenAlex bibliographic data; no institution, country, or ORCID provided) is an early-stage researcher working on full-duplex speech-to-speech models and conversational speech AI. The cataloged record centers on two 2026 first-author arXiv preprints: one enables streaming user transcription in a full-duplex speech-to-speech model via a parallel ASR head, reaching 10.21% WER with turn-taking behavior intact (streaming ASR, barge-in, and forced alignment are core keywords), and the other proposes a frontend-backend architecture for tool calls in full-duplex speech models. Evolve's verdict on the transcription work calls the method sensible but statistically thin and rates reproduction readiness medium. Bibliometrics are correspondingly narrow: 4 works in the last three years (1 in 2025, 3 in 2026), h-index 1, i10-index 0, and a single classified field, 'speech processing and conversational AI', yielding a maximal focus score of 100. Collaboration is the stronger signal — 21 distinct coauthors with 12 recurring names, including Edresson Casanova, Piotr Zelasko, and Boris Ginsburg — suggesting steady contribution within a larger speech team, for a collaboration score of 62. Topic data concentrates on Speech Recognition and Synthesis (4), speech and dialogue systems (3), and Speech and Audio Processing (2), with Topic Modeling (1) and Phonetics and Phonology Research (1) at the edges. Output and impact dimensions lack sufficient data to score.
Collaboration pattern
The record lists 21 distinct coauthors, with 12 recurring — including Edresson Casanova, Elena Rastorgueva, Nithin Rao Koluguri, Piotr Zelasko, Yifan Peng, Hainan Xu, Zhehuai Chen, Boris Ginsburg, and Slyne Deng — pointing to sustained work within a large, stable speech research group rather than loose one-off partnerships. Both catalog papers are first-authored but not corresponding, fitting a hands-on contributor role inside a team effort. Collaboration radar score: 62.
Research keywords
Highlights
Streaming user transcription for full-duplex speech-to-speech models
An honest read on the evidence: sensible method, thin statistics
A frontend-backend architecture for tool calls in full-duplex speech models
A 2026 first-author arXiv preprint proposes splitting tool-call handling in full-duplex speech models into frontend and backend components, extending duplex S2S systems beyond pure dialogue toward action-taking. Not yet reviewed by Evolve.
Timeline
- 2025Earliest indexed output: 1 work appears in the OpenAlex counts-by-year record.
- 2026Indexed output accelerates to 3 works, concentrated in speech processing and conversational AI.
- 2026First-author arXiv preprint enables streaming user transcription in full-duplex speech-to-speech models (10.21% WER, turn-taking intact).
- 2026First-author arXiv preprint proposes a frontend-backend architecture for tool calls in full-duplex speech models.
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.
- Name-resolved from public bibliographic data (OpenAlex); 'Ke Hu' is a common name and the record may merge or omit works. No institution, country, or ORCID is available.
- Citation impact cannot be assessed: h-index 1 and i10-index 0 are the only signals, and the radar leaves the output and impact dimensions unscored.
- The reproducibility score (55) is the mean over a single reviewed paper and is not a robust estimate.
- Field classification rests on one tagged paper ('speech processing and conversational AI'); the second catalog paper has no field, verdict, or slug yet.