Back to publications

Harf-Speech: A Clinically Aligned Framework for Arabic Phoneme-Level Speech Assessment

Asif Azad 1, MD Sadik Hossain Shanto 1, Mohammad Sadat Hossain 1, Bdour Alwuqaysi 1, Sabri Boughorbel 1, Yahya Bokhari 1, Abdulrhman Aljouie 1, Ayah Othman Sindi 2, Ehsan Hoque 1,3

  1. 1Ministry of Defense, Saudi Arabia
  2. 2Ability Center, Saudi Arabia
  3. 3University of Rochester, USA
Interspeech 2026 September 2026
First author ICORE A · 2026
Harf-Speech methodology from phoneme extraction through alignment and clinical scoring
Harf-Speech pipeline from phoneme extraction to interpretable word-level clinical scoring.

Abstract

Automated phoneme-level pronunciation assessment is vital for scalable speech therapy and language learning, yet validated tools for Arabic remain scarce. Harf-Speech is a modular system that scores Arabic pronunciation at the phoneme level on a clinical scale by combining an MSA phonetizer, a fine-tuned speech-to-phoneme model, Levenshtein alignment, and a blended scorer using longest common subsequence and edit-distance metrics.

The study fine-tunes three ASR architectures on Arabic phoneme data and benchmarks them against zero-shot multimodal models. The strongest model achieves an 8.92% phoneme error rate. Clinical validation by three certified speech-language pathologists shows a Pearson correlation of 0.791 and ICC(2,1) of 0.659 with mean expert scores, demonstrating clinically aligned and interpretable assessment comparable to inter-rater expert agreement.

Citation

Azad, A., Shanto, M. S. H., Hossain, M. S., Alwuqaysi, B., Boughorbel, S., Bokhari, Y., Aljouie, A., Sindi, A. O., & Hoque, E. (2026). Harf-Speech: A Clinically Aligned Framework for Arabic Phoneme-Level Speech Assessment. Accepted at Interspeech 2026.