Research on Deep Learning-Based English Speaking Automatic Scoring System

The application of deep learning technology in speech recognition and assessment provides innovative approaches for automatic English speaking scoring. This research designs a deep learning-based English speaking automatic scoring system, implementing intelligent assessment through advanced prosodic feature extraction, transformer-enhanced model architecture, and adaptive scoring mechanisms. At the technical level, methods such as comprehensive acoustic feature analysis, hybrid neural network construction with attention mechanisms, context-adaptive scoring rules, and real-time feedback integration are employed to enhance system fairness and diagnostic capability. Experimental results with diverse accent data show that the system achieves 89.3% scoring accuracy, with a correlation coefficient of 0.87 with human scoring, and system average response time below 200ms, providing an effective tool for English speaking teaching assessment.