Speech AI researcher advancing full-duplex speech-to-speech models with streaming transcription and tool calling
3
papers here
3
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
3
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 has been computed yet: this OpenAlex record is too sparse for a reliable productivity assessment beyond minimal raw counts (h-index 3, i10-index 2). The axis would measure publication volume, venue quality, and consistency of research output over time.
Impact—
No impact score is available yet: the record is early-stage and the 2026 arXiv preprints have had little time to accrue citations. The axis would measure citation counts, h-index trajectory, and downstream influence on subsequent speech-modeling research.
Focus8/100
The focus score reaches 100, and the raw data explains why: OpenAlex places this record in a single field, speech processing and conversational AI, and the topic profile is dominated by speech recognition and synthesis (10 works) and speech and dialogue systems (6), with smaller strands in topic modeling, phonetics and phonology, and speech and audio processing (3 each). Both catalogued 2026 papers target full-duplex speech-to-speech systems.
Collaboration86/100
The collaboration score of 62 rests on 21 distinct coauthors and a stable recurring core of 12 frequent collaborators, including Ke Hu, Edresson Casanova, Piotr Zelasko, Yifan Peng, and Boris Ginsburg. Both catalogued 2026 papers list this researcher in a middle author position without corresponding-author role, consistent with contributing inside larger speech teams rather than leading projects solo.
Activity61/100
The activity score of 67 is supported by an accelerating publication record: one indexed work in each of 2019, 2020 and 2021, then 2 in 2024, 3 in 2025 and 4 in 2026, totalling 9 works in the last three years. The 2026 output includes two reviewed full-duplex speech-to-speech papers, marking an intense current phase.
Reproducibility43/100
Reproducibility averages 55 across just one scored paper, so the figure is preliminary. Evolve's reviews of the two catalogued 2026 papers rate the delegation-token tool-calling work as low reproducibility readiness — strong benchmarks but no significance testing — while the streaming transcription paper reaches medium readiness with a reported 10.21% WER and thin statistics.
Research summary
Elena Rastorgueva is a speech-processing researcher whose work centers on full-duplex speech-to-speech models — systems that listen and speak simultaneously — with recent contributions on streaming user transcription and tool calling for voice agents. The OpenAlex bibliometric record is highly concentrated: all indexed activity falls in speech processing and conversational AI, with topic mass in speech recognition and synthesis (10 works) and speech and dialogue systems (6), plus smaller strands in topic modeling, phonetics and phonology, and speech and audio processing. Publication output has accelerated from single indexed works in 2019, 2020 and 2021 to 2 in 2024, 3 in 2025 and 4 in 2026 — 9 works in the last three years — alongside 21 distinct coauthors and a recurring core of 12 frequent collaborators. Two 2026 papers in the Evolve catalog define the current agenda: a frontend-backend delegation-token architecture that adds tool calls to a duplex speech frontend, and a parallel ASR head enabling streaming transcription at 10.21% WER with turn-taking intact. Evolve's reviews credit the engineering but flag thin statistical testing and mixed reproducibility readiness — low for the tool-call paper, medium for the transcription paper. Early-career indicators remain small (h-index 3, i10-index 2), and output and impact scores could not yet be computed.
Collaboration pattern
Elena Rastorgueva works inside sizeable speech teams: the record shows 21 distinct coauthors, with 12 frequent collaborators forming a stable core — among them Ke Hu, Edresson Casanova, Piotr Zelasko, Yifan Peng, and Boris Ginsburg. Both catalogued 2026 papers place this researcher in a middle author position without corresponding-author role, indicating contribution to shared system-building efforts rather than sole project leadership.
Research keywords
Highlights
Delegation-token architecture brings tool calling to full-duplex speech models
Streaming user transcription at 10.21% WER with turn-taking intact
A coherent research program on duplex spoken dialogue
The two reviewed 2026 papers attack one problem from two sides: keeping a duplex speech-to-speech model aware of what the user is saying while it speaks, and letting it act on that understanding through tool calls — evidence of a tightly focused, fast-moving agenda in conversational speech AI.
Timeline
- 2019First indexed work appears in the OpenAlex record (1 publication).
- 2020One further indexed work continues the early, still-sparse record.
- 2021One indexed work published; the last output before a visible gap in the record.
- 2024Publication activity resumes and picks up, with 2 indexed works in the year.
- 2025Output grows further to 3 indexed works.
- 2026Most active year so far: 4 indexed works, including two reviewed full-duplex speech-to-speech papers on tool calling and streaming transcription.
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.
- Bibliometrics are name-resolved from OpenAlex only; no ORCID, institution, or country is on file, so identity disambiguation relies solely on public bibliographic data.
- Output and impact radar scores are not computed, and citation coverage for the 2026 arXiv preprints is likely still incomplete.
- The reproducibility score (55) averages a single scored paper and should be read as preliminary.
- Only two papers carry full Evolve reviews; the other works counted in the yearly tallies are not individually reviewed.