Reproduce “AI-Driven Acoustic Voice Biomarker-Based Hierarchical Classification of Benign Laryngeal Voice Disorders from Sustained Vowels” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Mohsen Annabestani, Samira Aghadoost, Anais Rameau, Olivier Elemento, Gloria Chia-Yi Chiang

Researchers from Weill Cornell Medicine and Tehran University of Medical Sciences developed a three-stage hierarchical classification framework for eight benign laryngeal voice disorders and healthy controls using acoustic features from sustained vowels. This framework achieved an overall accuracy of 73.7% for 9-class subtype classification and outperformed general-purpose large speech models, demonstrating a clinically aligned approach for voice disorder diagnosis.

Key insights

  • A clinically inspired hierarchical classification framework significantly improves multi-class voice disorder diagnosis by breaking down the problem into sequential, manageable stages, mirroring medical reasoning.
  • Fusing deep spectral features with interpretable acoustic biomarkers provides a more robust and transparent diagnostic tool than either feature set alone or general-purpose pre-trained models.
  • Domain-specific AI models, designed for the nuances of sustained dysphonic vowels, consistently outperform generic large speech models (LSMs) that are pre-trained on conversational speech or event-driven health sounds.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.