The traditional English teaching mode generally consists of reciting words, phrases and texts with high intensity, mechanically memorizing grammar and doing a lot of exercises. This way not only causes the majority of students lack of practical ability to use English, but also difficult to stimulate students’ interest in learning. In this study, an immersive on-line teaching mode of audio-visual film English is constructed. Compared with the general on-line teaching mode, the model has a learning state feedback module based on convolutional neural network and support vector machine. Teachers can pay attention to students’ learning state in time and adjust teaching strategies according to their learning state. Three classes of the same school and grade were selected for comparative experiment. Offline teaching, traditional online teaching and improved online teaching were adopted in one semester respectively. The effectiveness of the improved method was proved by analyzing the class status and final grades of students in the whole semester. Through network training and practical tests, compared with the students who adopt traditional online teaching, the students who adopt this mode have greatly improved their learning state and final exam scores. At a time when the epidemic is still lingering, this mode can provide a positive reference for the development of online teaching.