Related Experiment Video
Updated: Jun 29, 2026
12:43
A Protocol for Comprehensive Assessment of Bulbar Dysfunction in Amyotrophic Lateral Sclerosis (ALS)
Published on: February 21, 2011
Attention-based Deep Feature Fusion for Automated Dysarthria Severity Classification: A Speech-based Computational
1Department of Electronics and Communication Engineering, Kumaraguru College of Technology, Coimbatore, 641049, India.
Current Neurovascular Research
|May 15, 2026
Summary
This study introduces an automated deep learning framework for classifying dysarthria severity. The attention-based fusion model achieved high accuracy, offering a robust tool for objective motor speech impairment assessment.
Area of Science:
- Neurology
- Speech Science
- Artificial Intelligence
Background:
- Dysarthria is a neuromotor disorder stemming from neurovascular and neurodegenerative diseases.
- It impairs motor speech control, leading to reduced intelligibility and communication challenges.
- Conditions like stroke and Parkinson's disease can cause dysarthria.
Purpose of the Study:
- To develop and evaluate a Deep Learning (DL) framework for automated dysarthria severity classification.
- To identify optimal DL methodologies for accurate dysarthria assessment.
- To provide an objective computational proxy for motor speech impairment analysis.
Main Methods:
- Investigated various DL models, including CNN and LSTM, with cepstral features.
- Utilized pretrained networks for feature extraction and hybrid DL-SVM approaches.
- Employed an attention-based fusion strategy on top-performing pretrained models.
Main Results:
- The attention-based fusion framework achieved high utterance-level accuracies (97.90% on TORGO, 95.31% on UA-Speech).
- Speaker-independent evaluations yielded lower but significant accuracies (62.77% and 56.26%).
- The proposed framework outperformed baseline methods in dysarthria severity classification.
Conclusions:
- Automated DL framework provides a robust, objective metric for assessing dysarthric speech.
- Pretrained networks combined with feature engineering optimize classification performance.
- The framework serves as a functional computational proxy for motor speech impairment analysis.