Mitigating Distractions in Virtual Classrooms: A Machine Learning Solution for Background Noise Detection

Virtual classrooms may encounter difficulties due to background noise. It can easily manage distracting noise that interferes with learning in a real classroom. In traditional techniques like spectrum subtraction and noise gating, dynamic and non-stationary background noise creates asymmetric patterns that are either suppressed too much or too little. Environments with non-stationary background noises, such as dog barking or overlapping speech, are difficult for such approaches to locate. It provides a machine learning (ML)-based approach for real-time background noise detection and filtering that uses supervised learning algorithms like Random Forest (RF) to distinguish between speech and noise in each audio segment. It provides outstanding accuracy and is characterized by the use of spectral centroid, ZCR (Zero-Cross Rate), and MFCCs (Mel-Frequency Cepstral Coefficients). When compared to the existing systems, the results indicate that the proposed system is superior with 94.5% accuracy, 93.2% precision, and 95.1% recall. Furthermore, it provides high voice preservation (low distortion) scores, making it appropriate for real-time online learning environments.