Steffen Freisinger
Georg Simon Ohm University of Applied Sciences Nuremberg
Speech processing researcher at Nuremberg working on ASR representation reuse and multimodal spoken-language corpora
2
papers here
2
reviewed by Evolve
1
first-author
0
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: publication counts and venue data in this snapshot are insufficient to compute a productivity measure. This axis would normally track the volume and consistency of peer-reviewed output over time.
Impact—
Impact is not yet scored: citation counts and influence indicators are absent from the available data, so no estimate is made. This axis would measure citations, downstream uptake, and scholarly influence of the reviewed work.
Focus0/100
The computed focus score is 0, driven by an even 0.5/0.5 split in the indexed field distribution between 'speech processing / automatic speech recognition' and 'speech processing and natural language processing'. In substance both labels lie within speech technology, so the low score reflects label breadth in the data rather than clearly divergent research directions; both reviewed papers sit on the ASR-NLP boundary.
Collaboration32/100
Collaboration scores 32. The snapshot records 4 distinct coauthors — Thomas Ranzenberger, Tobias Bocklet, Korbinian Riedhammer, and Philipp Seeberger — all recurring partners, suggesting a compact collaboration circle around the Nuremberg speech processing environment. Both reviewed papers are multi-author, with Freisinger first author on one and in a middle position on the other; neither lists him as corresponding author.
Activity—
Activity is not yet scored: the snapshot lacks year-by-year publication timing over a multi-year window, so research cadence cannot be computed. This axis would measure publication frequency and recent engagement, such as outputs per year around 2026.
Reproducibility55/100
Reproducibility scores 55, the mean over 2 scored papers, both rated 'medium' readiness. Evolve's verdicts flag concrete gaps: the first-author localization work reports consistent gains over text-only baselines but runs no significance tests, while the 553-hour corpus paper shows zero-shot slide context failing and leaves fine-tuning untested. Corpus-scale resources and strong baselines aid replication; unverified claims temper the score.
Research summary
Steffen Freisinger is a speech processing researcher at Georg Simon Ohm University of Applied Sciences Nuremberg whose reviewed work centers on automatic speech recognition and its interface with natural language processing. The Evolve catalog contains two of his 2026 arXiv papers. As first author he proposed reusing latent ASR encoder representations — Whisper states in particular — for query-conditioned topic localization in transcripts, reporting consistent gains over text-only baselines, though without significance testing and with gains that depend on domain. As a middle author he contributed to the Spoken Wikipedia Presentation Corpus, a 553-hour multimodal resource pairing Spoken Wikipedia audio with LLM-generated slides, where strong audio baselines were observed but zero-shot slide context did not help and fine-tuning remains untested. Bibliometric signals computed from public data show a narrow field distribution split evenly between speech processing/ASR and speech+NLP, four distinct coauthors, and a reproducibility score of 55 based on 2 papers. Collaboration scores 32 and the computed field-focus score is 0, reflecting the even split across the two indexed field labels; output, impact, and activity axes currently lack sufficient data and are intentionally left unscored.
Collaboration pattern
The recorded pattern is a small, stable circle: 4 distinct coauthors (Thomas Ranzenberger, Tobias Bocklet, Korbinian Riedhammer, Philipp Seeberger), all recurring, with a collaboration score of 32. Roles vary — first author on the query-conditioned localization paper, middle author on the corpus paper — and neither paper lists him as corresponding author. The pattern suggests close-knit team-based work typical of a university speech processing lab rather than wide external networking.
Research keywords
Highlights
Reusing ASR encoder states for query-conditioned topic localization
Co-building the 553-hour Spoken Wikipedia Presentation Corpus
Bridging speech representations and language tasks
Both reviewed papers sit at the ASR-NLP boundary: one reuses encoder states for text-side span localization, the other adds slide context to multimodal ASR. Together they outline an agenda of transferring speech-model internals into language-facing applications.
Timeline
- 2026Published first-author arXiv work reusing latent speech representations for query-conditioned topic localization in transcripts.
- 2026Co-authored the Spoken Wikipedia Presentation Corpus, a 553-hour multimodal ASR resource.
- 2026Bibliometric snapshot records 4 distinct coauthors and a reproducibility mean of 55 across the 2 reviewed papers.
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.
- Bibliometric scores are computed from OpenAlex-based public data only; output, impact, and activity axes lack sufficient data and are not scored.
- Coverage is limited to the two 2026 arXiv papers reviewed in the Evolve catalog; other publications, if any, are not reflected.
- No ORCID or country is linked in the source data; the profile is name-resolved from public bibliographic records.
- The reproducibility score (55) rests on only 2 papers, so the estimate is coarse.