Student or educator? Sign up with your school email and get 6 months of Pro free.

The DOI matches its Crossref record; 45 of the 45 references it cites were found in a registry or an index; 9 of the 44 sources its claims cite were checked for support.

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

Open access
Zhou, Zhou, Cheng, Yuan, Ren, Qingyi, Li, Yike, Yuanyue, Xu, Kang, Jing, Lu, Cheng, Ge, Pingjiang·2026·Frontiers in Digital Health

Abstract

Problem: Differentiating between spasmodic dysphonia (SD), a neurological disorder, and muscle tension dysphonia (MTD), a behavioral voice disorder, based on auditory perception alone is a common clinical challenge. This diagnostic difficulty can lead to delays in appropriate treatment, highlighting the need for objective and reliable assistive diagnostic tools. Aim: This study aims to develop and validate an artificial intelligence (AI) model based on deep learning to automatically differentiate between healthy voices, SD, and MTD using only voice audio recordings, and to compare its diagnostic performance against human experts. Methods: A retrospective analysis was conducted on 1,597 voice samples (595 healthy, 471 MTD, 531 SD). Voice audio was processed into Log-Mel spectrograms. Pre-trained convolutional neural networks (CNNs), including VGG16, ResNet50, and DenseNet161, were employed for transfer learning to perform both binary (Healthy vs. Disordered) and ternary (Healthy vs. MTD vs. SD) classification. The model's performance was evaluated on a held-out test set and compared to the diagnostic assessments of four otolaryngologists. Results: The AI model achieved an accuracy of 89.5% (AUC = 0.956) in distinguishing healthy voices from disordered ones. For the more complex ternary classification, the model attained an accuracy of 71.6%, with class-specific AUCs of 0.957 (Healthy), 0.731 (MTD), and 0.855 (SD). This performance surpassed that of human experts, who achieved average accuracies of 78.2% in binary classification and 60.6% in ternary classification on the same test set. Conclusion: The deep learning model trained on a Mandarin pathological voice dataset achieves favorable performance in distinguishing SD from MTD using only voice audio signals, with classification results comparable to those of experienced clinical specialists. This technology serves as a promising objective auxiliary tool for the preliminary screening and differential diagnosis of voice disorders.

Overview

CoChat Summary

TL;DR

This retrospective study trained deep learning models on 1,597 voice samples to distinguish healthy voices, muscle tension dysphonia (MTD), and spasmodic dysphonia (SD) from audio alone. The best model outperformed four human experts on both binary and three-class tasks, but the paper notes that its acoustic-only design is an initial screening tool rather than a full clinical diagnosis.

Key findings

  • On binary classification (healthy vs. disordered), the model achieved 89.5% accuracy, 92.2% sensitivity, 85% specificity, and AUC = 0.956.
  • On three-class classification (healthy vs. MTD vs. SD), the model achieved 71.6% accuracy, 70.2% sensitivity, 85.8% specificity, with class-specific AUCs of 0.957 (Healthy), 0.731 (MTD), and 0.855 (SD).
  • Human experts averaged 78.2% accuracy on the binary task and 60.6% on the three-class task; individual three-class accuracies were 59%, 60.5%, 59%, and 57.1%.
  • The paper reports the model inference was completed within seconds, whereas four specialists spent several hours evaluating the 324 test recordings.

Methods

  • Retrospective study using 1,597 voice samples: 595 healthy, 471 MTD, and 531 SD, collected from two hospitals in China from December 2012 to April 2024.
  • Audio was converted to Log-Mel spectrograms; three pre-trained CNNs (VGG16, ResNet50, DenseNet161) were fine-tuned with transfer learning and combined by averaging softmax probabilities.
  • Data were split patient-level into 960 training, 313 validation, and 324 test samples; performance was evaluated with AUROC, accuracy, sensitivity/recall, specificity, precision, F1-score, ICC, and McNemar’s test, and compared against four otolaryngology specialists.

Limitations

  • The paper states the sample is relatively small and limited to populations in southern China, which may restrict generalizability.
  • The model uses audio only; the paper says comprehensive voice-disorder diagnosis usually also requires medical history, symptoms, laryngoscopic findings, auditory perceptual assessment, and other information.
  • The paper notes it did not explicitly evaluate performance under real-world background noise or device variability; this is stated as a limitation and suggests robustness in deployment remains untested.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.3389/fdgth.2026.1690534

Cites

45 of 45 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Figure 1 An overview of the AI framework. Healthy = normal voice; MTD, muscle tension dysphonia; SD, spasmodic dysphonia.
Figure 1 An overview of the AI framework. Healthy = normal voice; MTD, muscle tension dysphonia; SD, spasmodic dysphonia.CC BY · PubMed Central
Figure 2 Confusion matrix of 2 and 3 classifications. Healthy = normal voice; MTD, muscle tension dysphonia; SD, spasmodic dysphonia.
Figure 2 Confusion matrix of 2 and 3 classifications. Healthy = normal voice; MTD, muscle tension dysphonia; SD, spasmodic dysphonia.CC BY · PubMed Central
Figure 3 Receiver operating characteristic curves of 2 and 3 classifications. Healthy = normal voice; MTD, muscle tension dysphonia; SD, spasmodic dysphonia.
Figure 3 Receiver operating characteristic curves of 2 and 3 classifications. Healthy = normal voice; MTD, muscle tension dysphonia; SD, spasmodic dysphonia.CC BY · PubMed Central
Figure 4 Confusion matrix of 3 classifications in human experts. Healthy = normal voice; MTD, muscle tension dysphonia; SD, spasmodic dysphonia.
Figure 4 Confusion matrix of 3 classifications in human experts. Healthy = normal voice; MTD, muscle tension dysphonia; SD, spasmodic dysphonia.CC BY · PubMed Central
Figure 5 Receiver operating characteristic plots for the benchmark tests. Clinical experts are marked by colored asterisks for individual performance and by an open circle for averaged performance. AUC, area under the curve.
Figure 5 Receiver operating characteristic plots for the benchmark tests. Clinical experts are marked by colored asterisks for individual performance and by an open circle for averaged performance. AUC, area under the curve.CC BY · PubMed Central
Figure 6 Examples of heat maps. Area marked by hot signals indicate the presence of graphic patterns contributing to a “positive” prediction. MTD, muscle tension dysphonia; SD, spasmodic dysphonia.
Figure 6 Examples of heat maps. Area marked by hot signals indicate the presence of graphic patterns contributing to a “positive” prediction. MTD, muscle tension dysphonia; SD, spasmodic dysphonia.CC BY · PubMed Central

Verification

Claims the paper makes

6 Valid · 1 Not valid · 10 Not checked · 64 Own results and interpretations

40 more claims

Create a free account to see all 81 claims and every source verdict.

See all claims — free account
  • It may state two separate claims as one. 26
  • Probably not a substantial claim. 19

Create a free account to see these claims.

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

2 of 2

Cited works that were retracted

Works this paper cites that were later retracted, from the retraction notices publishers deposit. Major when the notice came out before this paper was published.

1 of 1

Citations

Cites

1 retracted

Show all 44 references

Research works printed but not linked: 3

  1. Ludlow, 1995, Management of the spasmodic dysphonias, 436
  2. Roy, 2000, Toward a theory of the dispositional bases of functional dysphonia and vocal nodules: exploring the role of personality and emotional adjustment, 461
  3. Salhi, 2010, Voice disorders identification using multilayer neural network, Int Arab J Inf Technol, 2, 177

Metadata

Authors
Zhou, Z., Cheng, Y., Ren, Q., Li, Y., Yuanyue, X., Kang, J., Lu, C., & Ge, P.
Affiliations
  • Jinan University (CN)
  • Key Laboratory of Guangdong Province (CN)
  • Guangdong Academy of Medical Sciences (CN)
  • Vanderbilt University Medical Center (US)
  • Department of Otolaryngology Head Neck Surgery, Shenzhen People's Hospital (The Second Clinical Medical College, Jinan University, The First Affiliated Hospital of Southern University of Science and Technology)
Corresponding author
  • Kang, Jing
  • Lu, Cheng
  • Ge, Pingjiang
Venue
Frontiers in Digital Health
Published
2026-07-09
Volume
8
Pages
1690534-1690534
ISSN
2673-253X (ISSN-L)
OpenAlex
W7167808085
Cited by
1
References
44
Citations by year
2026: 1
FWCI
4.51
Type
journal article
Language
en
Open access
Locations
  • Frontiers in Digital Health (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • PubMed Central (publishedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion)
Funders
Basic and Applied Basic Research Foundation of Guangdong Province
Topics
  • Health Sciences › Health Professions › Speech and Hearing › Voice and Speech Disorders
  • Health Sciences › Health Professions › Health Information Management › Artificial Intelligence in Healthcare
  • Physical Sciences › Computer Science › Artificial Intelligence › Speech Recognition and Synthesis
Keywords
spasmodic dysphonia, muscle tension dysphonia, voice disorders, differential diagnosis, acoustic voice analysis, transfer learning, convolutional neural networks, log-mel spectrogram, pathological voice classification

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-01, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-01 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: A deep learning approach for acoustic-based identification of muscle tension dysphonia and spasmodic dysphonia. https://app.cochat.ai/papers/10.3389/fdgth.2026.1690534
The paper
Zhou, Z., Cheng, Y., Ren, Q., Li, Y., Yuanyue, X., Kang, J., Lu, C., & Ge, P. (2026). A deep learning approach for acoustic-based identification of muscle tension dysphonia and spasmodic dysphonia. Frontiers in Digital Health. https://doi.org/10.3389/fdgth.2026.1690534

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.