Research
Pashto speech and language resources, multilingual ASR evaluation, and patient-facing kidney disease education.
Public Google Scholar profile for Hanif Rahman, listed as independent researcher and verified at hanifrahman.com. Metrics were fetched on May 8, 2026.
Hanif Rahman
A reproducible public benchmark for Pashto automatic speech recognition across Whisper, MMS, SeamlessM4T, and OmniASR models. The study separates word error rate from script fidelity and documents cross-domain failure modes in read-speech datasets.
Hanif Rahman
Introduces Script Fidelity Rate, a reference-free metric for detecting speech recognition output in the wrong writing system. The benchmark covers ten languages, six scripts, and ten multilingual speech models.
Hanif Rahman
Compares full fine-tuning, LoRA, frozen encoders, and Urdu-to-Pashto transfer for Whisper on Pashto Common Voice. The paper reports practical scaling results and error patterns for Pashto-specific sounds.
Hanif Rahman, Shafeeq ur Rehman
Documents the growth of Pashto Common Voice from a small seed dataset into an openly licensed speech corpus with more than 100,000 clips. The work covers interface localisation, sentence collection, outreach, and baseline ASR results.
Hanif Rahman
Presents a 1.25-billion-word Pashto corpus, evaluation suite, and reproducible pipeline. The release includes data, model training, and code for low-resource NLP experiments.
Hanif Rahman
A comprehensive guide to IgA Nephropathy for patients, families, and anyone wanting to understand this common kidney disease. Covers diagnosis, progression, treatment options, and what to expect living with the condition.
Full citation list on Google Scholar