Uncovering Linguistic Patterns: Exploring Ensemble Learning and Low-Level Features for Identifying Spoken Arabic, English, Spanish, and German

This paper presents a novel approach to spoken language identification in Arabic, English, Spanish, and German languages using ensemble learning techniques. The study compares the performance of two well-known ensemble learning algorithms (Random Forest and XGBoost). Next, a Stacking Ensemble method with Logistic Regression is used as a meta-model, which combines the predictions of two configurations of Random Forest and two other configurations of XGBoost. We explore the utilization of the audio Low-Level Descriptors features, which have previously received limited attention in spoken language identification. Experimental results demonstrate the effectiveness of the ensemble learning algorithms, achieving high accuracy in accurately classifying spoken languages. Notably, the Stacking Ensemble method showcases its ability to reduce misclassification rate, emphasizing its potential for performance improvement. The stacking Ensemble method achieved the highest recorded classification rate of 97.22%, outperforming individual methods.