A machine learning-based intelligent evaluation and feedback system for college English speaking

To address the issues of insufficient evaluation frequency and unclear feedback orientation in college English oral language teaching, an intelligent oral language assessment and feedback system integrating machine learning was constructed.Based on real classroom speech data, multi-dimensional feature representations of pronunciation accuracy, speaking speed, pause ratio, and fluency were established.On this basis, a multi-model weighted scoring mechanism was formed.Experimental results showed that the system scoring was highly consistent with the manual evaluation, with a comprehensive accuracy rate of 0.84.The scoring deviation was concentrated within the ± 3-point range.After introducing a feedback adjustment mechanism based on learning records, the students' comprehensive scores increased by an average of 4.7 points within four weeks.Their speaking speed stabilised at 4.5-5.5 syllables per second, and the pause ratio decreased to approximately 0.21.The system is feasible in terms of process evaluation and teaching support, providing a data-driven auxiliary path for college English oral language teaching.