Related Experiment Video
Updated: May 10, 2026
12:43
A Protocol for Comprehensive Assessment of Bulbar Dysfunction in Amyotrophic Lateral Sclerosis ALS
Published on: February 21, 2011
34.9K
Early stroke diagnosis and evaluation based on pathological voice classification using speech enhancement.
Jun Zhang1, Yiyi Qiu1, Yingchen Liu1
1The State Key Laboratory of Digital Medical Engineering, Jiangsu Key Lab of Robot Sensor and Control, School of Instrument Science and Engineering, Southeast University, Nanjing, 210096, China.
Computers in Biology and Medicine
|August 17, 2025
Summary
This study developed an early stroke diagnosis (ESD) system using speech analysis and noise reduction. The system achieved 100% accuracy in clinical trials, offering a reliable tool for stroke detection.
Area of Science:
- Biomedical Engineering
- Artificial Intelligence
- Speech Processing
Background:
- Stroke prehospital screening tools are subjective and rely on medical knowledge.
- Speech analysis offers a non-invasive approach for early stroke diagnosis (ESD).
- Environmental noise significantly degrades the accuracy of speech-based diagnostic systems.
Purpose of the Study:
- To investigate the feasibility and effectiveness of ESD using pathological voice classification and speech enhancement (SE).
- To develop a cascaded framework for ESD integrating SE and recognition modules.
Main Methods:
- A SEWUNet-based SE module denoised sustained vowels (SVs) and spontaneous speech (SS).
- Recognition modules employed machine learning (KNN, SVM, RF, DT, AdaBoost) for SVs and CNN-Transformer/ResNet for SS.
- Five-fold cross-validation and incorporation of gender/age features were used for model training.
Main Results:
- Optimal SV models exceeded 90% accuracy, sensitivity, specificity, and F1-score.
- SS models surpassed 95% accuracy, with a 10% performance boost on enhanced speech.
- A real-time ESD system achieved 100% accuracy in clinical trials using a two-stage recognition strategy.
Conclusions:
- The proposed ESD method combining SE with SVs and SS serves as an assistive diagnostic tool.
- This approach aids in early stroke detection, reduces workload, and improves diagnostic objectivity.
- The study provides an open-source code and experimental protocol for ESD research.
