Standardized pronunciation assessment is crucial for English learners, especially given the inefficiencies of traditional methods. To address these challenges, this study develops an AI-based speech recognition and automatic feedback system aimed at improving the accuracy of oral error correction. This system is composed of two parts, namely, the voice recognition sensor and the neural network. Using Dynamic Time Warping (DTW) and Hidden Markov Model (HMM) to analyse the time sequence of voice signals, it can precisely detect voice mistakes and offer real time error-correcting feedback. The author used a standard speech corpus and scoring mapping model to construct an efficient scoring mechanism, significantly improving the reliability and intelligence level of speech processing. The experimental results show that when using this system to recognize selected natural speech segments, the recognition rate is higher than 94%, and the average video rate reaches over 96.5%; The recognition time fluctuates between 0.6 and 1.5 s, with an average recognition time of approximately 1.2 s. This system has significant advantages in error detection accuracy, feedback rationality, and user experience optimization. This study not only provides intelligent solutions for English oral teaching, but also opens up new directions for the application of speech recognition and artificial intelligence technology in the field of education. By integrating AI-driven speech recognition technology with a real-time feedback mechanism, users can enhance their English-speaking proficiency, providing valuable insights for the broader application of artificial intelligence in language education.