At present, online English listening and speaking teaching needs to strengthen the interaction between teachers and students, the interaction mode needs to be innovated, and the research level needs to be deepened. In order to extract the local and global information of the speech spectrum, better capture the spectrum structure of speech signals, and realize speech enhancement in low SNR and unfamiliar environments, this paper proposes a time-frequency domain speech enhancement network based on parallel multi-attention mechanism and codec structure. Experimental results show that compared with classic and mainstream models, the model proposed in this paper has achieved better enhancement effect. Moreover, the model proposed in this paper is effective in English listening and speaking teaching. Therefore, multi-interactive English teaching provides a new way of thinking for listening class, and it improves listening from the perspective of communication and interaction, and creates a real scene of listening and speaking to improve students’ comprehensive English application ability.