Speech emotion recognition is the basis of emotional speech synthesis, a good speech emotion recognition system can learn more emotional expressions in speech and help in the synthesis of emotional speech. However, there are a number of issues that make the speech emotion recognition task difficult, including background noise and the distinct speech features of each speaker. The widely recognized speech emotion recognition system ACRNN extracts local features from speech signals using CNN, and its attention mechanism concentrates on the emotional content of the speech data. However, because only a single attention module is used, it is unable to simultaneously attend to the information from distinct representation subspaces at different locations, nor is it able to acquire long-term global information. The paper proposes CoRNN, which applies Conformer to replace CNN and attention module, with the purpose of overcoming the shortcomings of ACRNN. The experimental results on IEMOCAP dataset demonstrate that the unweighted average recall of the proposed CoRNN can achieve 65.53%, which improves 0.79% comparing with ACRNN.