With the emergence of deep learning technologies, speech enhancement methods based on deep learning have seen widespread application and generally surpass traditional approaches in performance. The fundamental framework of noise reduction signal processing in speech enhancement was outlined and progressively delved into the latest advancements in deep learning-driven speech enhancement models. A comprehensive organization of deep learning-based speech enhancement algorithms was provided, detailing the principles, characteristics, evaluation metrics, and representative studies of various neural network-based methods. The advantages and limitations of these approaches were thoroughly assessed. Finally, in light of the current developmental landscape, the core challenges encountered in the speech enhancement process were analyzed, and future developmental trajectories were discussed and predicted.