Mizbaul Haque Maruf

Islamic University of Technology · Bangladesh

Unclaimed · AI-structured profileOpenAlex

Bangla low-resource speech AI researcher — corpora, TTS and turn-taking benchmarks, Islamic University of Technology

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-10-13) 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 scored: publication-volume data is insufficient in this OpenAlex record. This axis would measure total research output — the number of indexed works, their venues and types, and productivity over time — none of which can be responsibly estimated from the current sparse record.

Impact—

Not yet scored: citation signals are too sparse at this stage (h-index 1, i10-index 1). This axis would measure citation-based influence — citation counts, h-index trajectory, and uptake of the released Bangla corpora and models by other research groups.

Focus0/100

Scores 0/100 because the field distribution splits evenly — about 33% each — across three closely related labels: speech and language processing, low-resource TTS, and computational linguistics for TTS. Topic counts (speech recognition and synthesis: 3; phonetics and phonology: 2; AI in education: 2) suggest overlapping labeling rather than genuinely scattered work; the de facto focus is Bangla speech technology.

Collaboration22/100

Scores 22/100. OpenAlex records only 2 distinct co-authors in the recent three-year window, and the catalog lists Muhammad Nur Yanhaona as the recurring partner. All three reviewed papers are first-authored, two with corresponding-author role, indicating a small-team, lead-author working mode typical of an early-stage researcher building an independent agenda rather than operating inside a large lab network.

Activity52/100

Scores 52/100 on 5 works in the recent three-year window. Yearly counts — 1 in 2023, 1 in 2025, and 4 in 2026 — show a sharp recent acceleration, with 2026 alone supplying the three first-author arXiv preprints reviewed by Evolve. The momentum is real but the base is thin, so sustained consistency across years is still unproven.

Reproducibility70/100

Scores 70/100, the mean readiness rating across the 3 Evolve-reviewed papers (reproMean 70 over reproPapers 3). BanglaKontho rates high on the strength of a documented 20-hour corpus with careful QA; BanglaTurn is medium, with its honestly stated single-annotator limitation; Structure Before Sampling is medium despite solid null tests. Resource-first releases with explicit limitations support this above-average signal.

Research summary

Mizbaul Haque Maruf is an early-career researcher at the Islamic University of Technology in Bangladesh working on low-resource Bangla speech technology. This profile is name-resolved from public OpenAlex bibliographic data, which records an h-index of 1 and five works in the last three years, with output concentrated in 2026 (four works, including the three first-author arXiv preprints reviewed by Evolve). Two are corresponding-author resource releases: BanglaTurn, the first public Bangla turn-end detection benchmark, where a Whisper-based model outperforms Smart-Turn v3 by 15 percentage points on a 319-clip test set; and BanglaKontho, a 20-hour single-speaker audiobook text-to-speech corpus built with careful quality assurance. A third paper introduces community-aware core-set selection for data-efficient TTS, beating random and entropy baselines at equal budgets. Evolve's reproducibility screening averages 70 across the three papers, with BanglaKontho rated high. Topic clustering places the work in speech recognition and synthesis, phonetics and phonology, and AI in education. Output and impact dimensions are not yet scored because citation data is too sparse, and the record shows only two distinct co-authors, with Muhammad Nur Yanhaona the recurring collaborator — an early-stage profile built on resource release rather than citation accumulation.

Collaboration pattern

The collaboration footprint is narrow: the recent three-year OpenAlex window records only 2 distinct co-authors, matching a collaboration score of 22. Muhammad Nur Yanhaona is the recurring catalog co-author. All three reviewed papers are first-authored by Maruf, two also corresponding, suggesting a small-team mode in which he leads corpus building, modeling, and write-up himself.

Research keywords

end-of-turn detectionturn-takingBangla speech processinglow-resource languagesWhisperBangla text-to-speechspeech corpus constructionaudiobook prosodytext normalizationcore-set selectiondata-efficient trainingcommunity detection

Highlights

BanglaTurn: the first public Bangla end-of-turn detection benchmark
Releases the first public Bangla turn-end corpus and a Whisper-based detector that outperforms Smart-Turn v3 by 15 percentage points on a 319-clip test set. Evolve credits the work's honesty but flags reliance on a single annotator as the main limit on certainty.
BanglaKontho closes the long-form gap in Bangla text-to-speech
Releases a 20-hour single-speaker Bangla audiobook TTS corpus with careful quality assurance, targeting long-form synthesis. Evolve judges the gains over IndicTTS-Bn credible while noting the domain-versus-scale attribution remains untested; the paper carries a high reproducibility-readiness rating.
Community-aware core-set selection for data-efficient TTS
Proposes selecting TTS training data via structure on a phonotactic graph using community detection and Louvain modularity, before any sampling. It beats random and entropy baselines at equal data budgets with solid null tests, though ablations and speaker scope are thin.

Timeline

2023First indexed work appears in the OpenAlex yearly counts (1 work); it is not part of the Evolve catalog and is not described further.
2025One indexed work in 2025 — a sparse year between the 2023 entry and the 2026 output peak.
2026Most productive year: 4 indexed works, including three first-author arXiv preprints reviewed by Evolve — BanglaTurn, BanglaKontho, and Structure Before Sampling.

Papers in the Evolve catalog

2026-09-24first authorcorresponding
BanglaTurn: A Benchmark and Whisper-Based Model for End-of-Turn Detection in Bangla Speech
2026-09-24first authorcorresponding
BanglaKontho: Closing the Long-Form Gap in Bangla Text-to-Speech
2026-09-21first author
Structure Before Sampling: Community-Aware Core-Set Selection for Data-Efficient Text-to-Speech

Frequent co-authors

Mizbaul Haque Maruf×1 Muhammad Nur Yanhaona×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.

Bibliometric indicators (h-index, yearly counts, topics) come from OpenAlex only; no ORCID was available, so identity is resolved from public bibliographic data.
Output and impact dimensions are unscored: citation signals are too sparse at this stage to support any productivity or influence estimate.
Only the three Evolve-reviewed arXiv preprints are assessed in depth; the 2023 and 2025 indexed works are outside the catalog and are not described.
The reproducibility score reflects Evolve's readiness screening (mean 70 over 3 papers), not independent replication of results.