Voice Disorder Classification Using Wav2vec 2.0 Feature Extraction.

Area of Science:

• Speech processing and biomedical signal analysis.
• Application of deep learning and machine learning in healthcare.

Background:

• Accurate voice classification is crucial for diagnosing various pathological conditions.
• Traditional methods may lack the feature extraction capabilities needed for complex voice patterns.

Purpose of the Study:

• To classify normal versus pathological voices using wav2vec 2.0 for feature extraction.
• To evaluate the performance of different machine learning classifiers in this task.

Main Methods:

• Utilized the wav2vec 2.0 model for extracting features from voice recordings.
• Trained and evaluated Support Vector Machine (SVM), K-Nearest Neighbors, Decision Tree (DT), and Random Forest (RF) models.
• Employed Stratified K-Fold cross-validation and various performance metrics including accuracy, precision, recall, F1-score, ROC curve, and confusion matrix.

Main Results:

• The Random Forest (RF) model achieved the highest accuracy (0.98 ± 0.02) and strong performance across other metrics.
• Decision Tree (DT) model showed excellent precision and balanced performance.
• Data augmentation significantly improved the performance of all models, particularly RF and DT.

Conclusions:

• The combination of wav2vec 2.0 and machine learning models is highly effective for voice classification.
• Achieved exceptional accuracy and robustness in distinguishing normal from pathological voices.
• RF and DT models are particularly well-suited for this voice classification task.