King's College London researcher interpreting speaker embeddings with hierarchical clustering and explainable AI.
2
papers here
2
reviewed by Evolve
0
first-author
0
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 metrics are available yet. This axis would measure publication volume and trajectory — e.g. yearly paper counts and output consistency — but the current snapshot contains only the two reviewed 2026 arXiv papers, so no score is given.
Impact—
No citation or impact metrics are available in the current data. This axis would measure scholarly influence — citations, citations-per-paper and field-normalised impact — but with only two 2026 arXiv papers, no reliable score can be computed.
Focus0/100
The radar focus score is 0, with the field distribution split evenly at 0.5 across two adjacent areas: 'speech processing / speaker recognition (explainable AI)' and 'speech technology and machine learning (explainable AI)'. Both 2026 papers sit in the same explainable-AI-for-speaker-embeddings niche, so the low score reflects the even two-field split in the bibliometric classification rather than visible topic dispersion across the reviewed work.
Collaboration22/100
The collaboration score is 22, based on only 2 distinct coauthors in the sample. The recurring partners are Yanze Xu and Wenwu Wang, and Plumbley appears in the last author position on both reviewed 2026 papers without being corresponding author — a compact, stable team pattern consistent with a senior supervisory role rather than a large distributed network. The small coauthor pool makes this score highly sensitive to sample size.
Activity—
No activity data is available yet. This axis would measure ongoing research engagement — recent publication cadence, preprint activity and continuity over time — but the snapshot lacks the longitudinal record needed to score it.
Reproducibility38/100
The reproducibility score is 38, a mean computed over 2 assessed papers. Evolve rates the HCCM/L-score speaker-embedding paper as medium readiness, while the second-order pattern paper is low readiness, its review noting that the Gradient Cost measure is heuristic and the evaluation thin. Together the two ratings place reproducibility in the lower-middle band, indicating methods that are documented but not yet easy to replicate independently.
Research summary
Mark D. Plumbley is a researcher at King's College London whose reviewed work in the Evolve catalog centres on making speaker embeddings interpretable through hierarchical clustering and explainable AI. Two arXiv papers, both dated 2026, define a coherent single-thread agenda. The first introduces HCCM and an L-score for interpreting SLINK single-linkage hierarchies of speaker embeddings; Evolve's review finds the findings interesting but notes a single model and dataset and no clustering baselines. The second extends the same SLINK-plus-HCCM toolkit with a new HCNA method to discover and assign latent second-order patterns in speaker recognition, though the review flags the Gradient Cost measure as heuristic and the evaluation as thin. On both papers Plumbley appears in the last author position without being corresponding author, working with a compact set of 2 distinct coauthors in the sample — frequent partners Yanze Xu and Wenwu Wang. Radar scores reflect this snapshot: collaboration 22, focus 0 (field distribution split 0.5/0.5 across two adjacent speech-and-machine-learning areas) and reproducibility 38 averaged over 2 papers. Output, impact and activity axes lack sufficient data. This is a full but early-coverage profile built from a small, 2026-only bibliographic sample.
Collaboration pattern
The sampled collaboration footprint is compact: 2 distinct coauthors, with Yanze Xu and Wenwu Wang as recurring partners. On both reviewed 2026 arXiv papers Plumbley sits in the last author position without being the corresponding author, a pattern typical of a senior researcher supervising a close-knit team rather than coordinating a wide network. The radar collaboration score of 22 reflects this small pool and is sensitive to the limited 2-paper sample.
Research keywords
Highlights
HCCM and L-score: interpreting SLINK hierarchies of speaker embeddings
HCNA uncovers latent second-order patterns in speaker embeddings
A one-year, single-thread explainability program in speaker recognition
Both reviewed papers appeared in 2026 and share SLINK, HCCM and the same embedding-analysis target, forming a focused explainable-AI line on speaker embeddings. Both carry a reproducibility mean of 38 across the 2 papers (medium and low readiness) and place Plumbley in the last-author seat with only 2 distinct coauthors in the sample.
Timeline
- 2026Publishes 'Interpreting hierarchical organisation of speaker embeddings' on arXiv as last author, introducing HCCM and the L-score for interpreting SLINK hierarchies.
- 2026Publishes 'Exploring Second-Order Pattern Recognition in Speaker Recognition' on arXiv, extending SLINK+HCCM with the new HCNA method, again as last author.
- 2026The two 2026 papers consolidate one explainable-AI line on speaker embeddings, with a reproducibility mean of 38 across the 2 assessed 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 snapshot is small: radar scores derive from only 2 reviewed 2026 arXiv papers and 2 distinct coauthors, so scores are highly sample-sensitive.
- Output, impact and activity axes are marked missing — no citation counts, publication-volume or longitudinal activity data were available.
- Catalog coverage reflects Evolve-reviewed work only; older or non-arXiv publications are not represented in this profile.
- Profile is name-resolved from public bibliographic data under ORCID 0000-0002-9708-1075; no biographical details beyond the listed affiliation are verified.