An enhanced model for environmental sound classification using bio-inspired multi-kernel optimization algorithm

Environmental sound classification is a rapidly growing research area with numerous applications. However, current models frequently face challenges in accurately identifying subtle patterns in environmental sounds due to inherent noise and variability, leading to reduced efficiency and performance. To address these challenges, we propose a novel approach that enhances feature selection and kernel optimization through an enhanced elitism-based Grey Wolf Optimization algorithm (Env-GWO). This method identifies optimal features and uses multiple kernel weights to capture diverse aspects of sound data, reducing the risk of local optima and increasing robustness to noise and variability. Our contributions are twofold: firstly, the enhanced feature selection uses Env-GWO to explore a wide range of feature combinations, significantly improving the ESC model's performance by overcoming the limitations of conventional methods in noisy conditions. Secondly, the improved ESC model incorporates Env-GWO for feature selection and kernel optimization, capturing different aspects of environmental sounds through multiple sets of kernel weights, thereby increasing the model's adaptability to varying noise levels and complex acoustic environments. Experimental results demonstrate that our multi-solution approach advances the field of environmental sound analysis, achieving a more comprehensive and accurate representation of environmental sound data.