Analysis and Classification of Voice Pathologies Using Glottal Signal Parameters

Area of Science:

• Medical Engineering
• Speech Pathology
• Artificial Intelligence in Healthcare

Background:

• Accurate voice disease classification is crucial for effective treatment and medical device design.
• Distinguishing between vocal fold pathologies like nodules and paralysis is clinically significant.
• Glottal signal analysis offers a promising avenue for objective voice disorder assessment.

Purpose of the Study:

• To classify voice signals into three categories: vocal fold nodule, unilateral paralysis, and normal voice.
• To evaluate the effectiveness of Artificial Neural Networks (ANN), Support Vector Machines (SVM), and Hidden Markov Models (HMM) in voice disorder classification.
• To compare the classification performance of these models using glottal signal parameters.

Main Methods:

• Extracted glottal signal parameters using inverse filtering.
• Utilized ANN, SVM, and HMM for voice signal classification.
• Trained and tested models on a database of 248 voice recordings from three distinct groups.

Main Results:

• Achieved a superior classification rate of 97.2%, outperforming previous studies.
• Demonstrated the efficacy of the selected machine learning models in differentiating voice disorders.
• The study utilized a larger database compared to similar research.

Conclusions:

• Glottal signal parameters effectively support the identification of vocal fold nodules and unilateral paralysis.
• Machine learning models, particularly ANN, SVM, and HMM, are powerful tools for voice disorder classification.
• The findings contribute to improved diagnostic capabilities for voice pathologies.