A deep learning approach for acoustic-based identification of muscle tension dysphonia and spasmodic dysphonia - Report - MDSpire

A deep learning approach for acoustic-based identification of muscle tension dysphonia and spasmodic dysphonia

  • By

  • Zhou Zhou

  • Yuan Cheng

  • Qingyi Ren

  • Yike Li

  • Xu Yuanyue

  • Jing Kang

  • Cheng Lu

  • Pingjiang Ge

  • July 9, 2026

  • 0 min

Share

Clinical Report: Utilizing Deep Learning for Acoustic Identification of Dysphonia

Overview

This study developed a deep learning model to differentiate between healthy voices, spasmodic dysphonia (SD), and muscle tension dysphonia (MTD) using voice audio recordings. The AI model achieved 89.5% accuracy in binary classification and 71.6% in ternary classification.

Background

Differentiating between SD and MTD is clinically challenging due to their overlapping perceptual features and similar laryngoscopic findings. Misdiagnosis can lead to inappropriate treatment, emphasizing the need for objective diagnostic tools. This study addresses the gap in existing frameworks by implementing a three-class classification system for these voice disorders.

Data Highlights

ClassificationAccuracyAUC
Binary (Healthy vs. Disordered)89.5%0.956
Ternary (Healthy vs. MTD vs. SD)71.6%
Class-specific AUC (Healthy)0.957
Class-specific AUC (MTD)0.731
Class-specific AUC (SD)0.855

Key Findings

  • The AI model achieved 89.5% accuracy in distinguishing healthy voices from disordered ones.
  • For ternary classification, the model attained 71.6% accuracy.
  • Class-specific AUCs were 0.957 for healthy voices, 0.731 for MTD, and 0.855 for SD.
  • Human experts achieved average accuracies of 78.2% in binary and 60.6% in ternary classifications.
  • The model serves both screening and differential diagnostic functions.

Clinical Implications

The AI model provides an objective tool for the preliminary screening and differential diagnosis of voice disorders.

Conclusion

The deep learning model demonstrates performance in distinguishing between SD and MTD.

Related Resources & Content

  1. Clinical Practice Guideline: Hoarseness (Dysphonia) (Update) - Stachler - 2018 - Otolaryngology–Head and Neck Surgery
  2. Position Statement: Botulinum Toxin Treatment - American Academy of Otolaryngology-Head and Neck Surgery (AAO-HNS)
  3. Acoustic Metrics of the Strain Dimension of Voice Quality: A Scoping Review
  4. Analysis of Acoustic Signals in Instrument-Tissue Interactions During Minimally Invasive Procedures
  5. Frontiers in Medicine — Improving respiratory disease detection through SSL-enhanced acoustic analysis and exercise-rest measurements
  6. npj Digital Medicine — A device-invariant multi-modal learning framework for respiratory disease classification
  7. Frontiers in Digital Health — Data-driven refinements for voice disorder classification: improving accuracy and generalisability
  8. Clinical Practice Guideline: Hoarseness (Dysphonia) (Update) - Stachler - 2018 - Otolaryngology–Head and Neck Surgery - Wiley Online Library
  9. Position Statement: Botulinum Toxin Treatment - American Academy of Otolaryngology-Head and Neck Surgery (AAO-HNS)
  10. Acoustic Metrics of the Strain Dimension of Voice Quality: A Scoping Review - ScienceDirect

Original Source(s)

A deep learning approach for acoustic-based identification of muscle tension dysphonia and spasmodic dysphonia

  • by Zhou Zhou, Yuan Cheng, Qingyi Ren, Yike Li, Xu Yuanyue, Jing Kang, Cheng Lu, Pingjiang Ge

  • July 9, 2026

Related Content

Anatomy-guided dissection plane selection in endoscopic brow lift: a narrative review of supraperiosteal, subperiosteal, and multiplane approaches

Target-prioritized IMRT for nasopharyngeal carcinoma with tumor proximity to the spinal cord: clinical feasibility and long-term outcomes

Longitudinal voice biomarker trajectory modelling for Parkinson's disease severity: domain-adaptive transfer learning on mPower real-world smartphone data