Feed-Forward Deep Neural Network Model Based Speech Recognition System for Speech Signal

This research work aims to enhance speech recognition accuracy and system generalization performance by optimizing deep neural network (DNN) Systems. The Experiments are conducted using a standard benchmark speech dataset and an independent real-time speech dataset while following a complete speaker-independent assessment method. The baseline model uses a feed-forward DNN, which researchers improve through Genetic Algorithm (GA) and Particle Swarm Optimization (PSO) and Whale Optimization Algorithm (WOA), and the proposed Neural Whale Optimization Algorithm (NOWOA). Comprehensive evaluations, including confusion matrix-based metrics, 5-fold cross-validation, and overfitting analysis, are performed to assess robustness and reliability. Experimental results demonstrate that the baseline DNN achieves approximately 50\% recognition accuracy, while optimization significantly enhances performance. The proposed NOWOA-optimized DNN system achieves the highest recognition accuracy of 99.36\% among all tested methods, proving its effectiveness for speech recognition tasks on both standard and real-time datasets.