This study applies deep learning techniques to improve speech recognition in different environments, including noisy situations. The study presents the results of a Convolutional Neural Network (CNN) model that achieved high accuracy in recognizing spoken words in clean audio data but struggled as noise levels increased. An attempt to use a simple autoencoder to remove noise from the audio data resulted in a significant decline in classification performance, indicating the need for more advanced denoising techniques. The study proposes the Speech enhancement Generative Adversarial Network (SEGAN) as a solution to improve speech recognition's robustness by generating clean speech signals from noisy inputs. The results demonstrate the potential of deep learning techniques in improving speech recognition technology, but careful evaluation of various components and techniques is crucial for optimal performance.
