Аңдатпа
This study is designed to investigate effective methods for artificial intelligence-based detection of hyperkinetic speech disorders using audio signals. The main objective of the study is to compare several machine learning models and create a platform for automatic diagnosis of speech disorders using the Hyperkinetic Dysarthria Speech (HDS) dataset. The data was pre-processed and acoustic features (MFCC, Chroma, Spectral Centroid, Zero-Crossing Rate) were extracted. The models compared included Logistic Regression, K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Decision Tree, Random Forest, Gradient Boosting, AdaBoost, Extra Trees, XGBoost, LightGBM, CatBoost, and Multi-Layer Perceptron Classifier (MLPClassifier). According to the results, the Logistic Regression, KNN, SVM, CatBoost, and MLPClassifier models reached the level of 1.0 in terms of Accuracy, Precision, Recall, and F1-Score metrics, which indicates that pathological features in speech signals are classified with high accuracy. The research resulted in the development of the Streamlit platform based on the MLPClassifier model, which allows users to upload audio files or record speech in real time. This system reduces the subjectivity of traditional clinical assessments, facilitates early diagnosis, and helps improve the quality of life of patients. However, limitations of the study include the availability of data mainly in Russian and the relatively small size of the dataset (approximately 760 files), which limits the generalizability of the models to other languages. Also, the stability in noisy environments requires additional testing. Future studies are planned to incorporate multilingual data, use deep learning architectures, and validate the platform in clinical settings. This work will strengthen the role of computational hearing and machine learning in the detection of speech pathology and contribute to innovations in the field of neurological diagnostics and therapy.