This paper explores the development of an intelligent translation system for spoken English using Recurrent Neural Network (RNN) models. The fundamental principles of RNNs and their advantages in processing sequential data, particularly in handling time-dependent natural language data, are discussed. The methodology for constructing the translation system is outlined, covering key steps such as data preprocessing, model architecture design, and training optimization. The system's performance is evaluated in terms of translation accuracy, fluency, and real-time processing capabilities. The study identifies limitations of the current system and proposes future research directions, including the integration of attention mechanisms, refinement of model architectures, and enhancement of multilingual translation capabilities. Ultimately, this research contributes theoretical insights and practical guidance to the ongoing development of intelligent translation systems for spoken English.