Tobias Bocklet

Georg Simon Ohm University of Applied Sciences Nuremberg

Unclaimed · AI-structured profileORCID OpenAlex

Speech processing researcher (Ohm Nuremberg) — multimodal ASR, latent speech representations, spoken-language resources

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—

Output is not yet scored: this axis would measure publication volume and productivity over time — papers per year, venue spread, authorship roles — but the catalog holds only two 2026 arXiv entries, too few for a reliable score.

Impact—

Impact is not yet scored: this axis would track citations, downstream reuse, and scholarly influence of the work, but no citation data exists in the current snapshot, so no estimate is offered.

Focus0/100

The computed focus score is 0, driven by a field distribution split exactly 50/50 between 'speech processing / automatic speech recognition' and 'speech processing and natural language processing' — the two catalog papers map one-to-one onto these labels. In practice the topical spread is narrow: one paper reuses Whisper encoder representations for query-conditioned topic localization in transcripts, the other contributes a 553-hour multimodal ASR corpus with slide context, both within the speech-and-language area.

Collaboration32/100

The collaboration score is 32/100, computed from 4 distinct coauthors across the two catalog papers. On both papers he is a middle author and not the corresponding author, indicating contribution within a team rather than independent leadership. The recurring partners — Thomas Ranzenberger, Steffen Freisinger, Korbinian Riedhammer, and Philipp Seeberger — point to a small, stable working group rather than a broad external network.

Activity—

Activity is not yet scored: this axis would measure recent publishing cadence and continuity across years, but with only two 2026 papers on record there is no multi-year signal to evaluate.

Reproducibility55/100

The reproducibility mean is 55, averaged over 2 assessed papers, each rated medium readiness. Evolve's verdicts align with that: the topic-localization paper reports consistent gains over text-only baselines without significance tests, while the corpus paper shows strong audio baselines but a failing zero-shot slide-context condition and untested fine-tuning, leaving the claimed gain unproven.

Research summary

Tobias Bocklet is a speech-technology researcher at Georg Simon Ohm University of Applied Sciences Nuremberg whose recorded output sits at the intersection of automatic speech recognition and natural language processing. The Evolve catalog contains two of his 2026 arXiv papers, in both of which he appears as a middle author rather than corresponding author. The first, 'Reusing Latent Speech Representations for Query-Conditioned Topic Localization in Transcripts', reuses ASR encoder states built on Whisper to localize query-conditioned topic spans in transcripts for topic segmentation and spoken document retrieval; Evolve's review notes consistent gains over text-only baselines, but no significance tests and domain-dependent gains. The second, 'The Spoken Wikipedia Presentation Corpus', introduces a 553-hour resource pairing Spoken Wikipedia audio with LLM-generated slides for multimodal ASR and contextual biasing; audio baselines are strong, while zero-shot slide context fails and fine-tuning remains untested. His radar profile shows a field distribution split evenly (0.5/0.5) between 'speech processing / automatic speech recognition' and 'speech processing and natural language processing', a collaboration score of 32 built on 4 distinct coauthors, and a reproducibility mean of 55 across 2 assessed papers, both rated medium. Output, impact, and activity axes are not yet computable from the available data. He works within a stable small team: Thomas Ranzenberger, Steffen Freisinger, Korbinian Riedhammer, and Philipp Seeberger.

Collaboration pattern

Tobias Bocklet's collaborations form a compact cluster: 4 distinct coauthors across two 2026 catalog papers, with a recurring core of Thomas Ranzenberger, Steffen Freisinger, Korbinian Riedhammer, and Philipp Seeberger. On both papers he is a middle author and not the corresponding author, consistent with contributing inside a stable speech-processing group at Georg Simon Ohm University of Applied Sciences Nuremberg rather than leading independently. The low collaboration score (32/100) reflects this small, concentrated pattern.

Research keywords

automatic speech recognitionASR encoder representationsquery-conditioned span localizationWhispermultimodal fusionmultimodal ASRcontextual biasingtopic segmentationspoken document retrievalspoken language modelscorpus construction

Highlights

Query-conditioned topic localization by reusing ASR encoder states
In a 2026 arXiv paper (middle author), latent speech representations from the ASR encoder — Whisper-based — are reused to localize query-conditioned topic spans in transcripts, supporting topic segmentation and spoken document retrieval. Evolve's review records consistent gains over text-only baselines, while noting the absence of significance tests and domain-dependent gains.
Co-building the 553-hour Spoken Wikipedia Presentation Corpus
A second 2026 arXiv release (middle author) introduces SWPC, a 553-hour resource pairing Spoken Wikipedia audio with LLM-generated slides to support multimodal ASR and contextual-biasing research. Audio baselines are strong; zero-shot slide context fails and fine-tuning remains untested, so the intended gain is not yet proven.
Candid limitation reporting alongside new resources
Both catalog entries pair new methods or resources with openly reported limitations — no significance tests in the localization study, and a failing zero-shot slide-context condition in the corpus paper — a pattern consistent with the medium (55) reproducibility rating across the 2 assessed papers.

Timeline

2026Co-authored (middle position) 'Reusing Latent Speech Representations for Query-Conditioned Topic Localization in Transcripts' on arXiv, reusing Whisper-based encoder states for query-conditioned topic localization in transcripts.
2026Co-authored (middle position) 'The Spoken Wikipedia Presentation Corpus' on arXiv — a 553-hour resource pairing Spoken Wikipedia audio with LLM-generated slides for multimodal ASR research.
2026Both 2026 papers entered the Evolve catalog with medium reproducibility readiness (repro mean 55 across 2 assessed papers); output, impact, and activity scores remain unavailable at this early stage.

Papers in the Evolve catalog

Frequent co-authors

Thomas Ranzenberger×2 Steffen Freisinger×2 Korbinian Riedhammer×2 Philipp Seeberger×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 come from the OpenAlex-based Evolve snapshot only; citation counts and venue histories are outside the current data scope.
Only two papers, both 2026 arXiv preprints, are in scope; peer-reviewed publications and pre-2026 work are not reflected.
Output, impact, and activity axes are marked missing: the underlying data is insufficient and no scores or estimates are offered.
Country metadata is not provided; this profile is name-resolved via ORCID 0009-0008-7780-8821 from public bibliographic data.