Modern Automatic Speech Recognition (ASR) systems aim to accurately convert spoken language into written text. However, they often face challenges when confronted with emotional speech, as traditional systems struggle to interpret the subtleties of emotional inflection. To overcome this challenge, a multi-task learning approach has been proposed that simultaneously addresses ASR and Speech Emotion Recognition (SER). With limited emotional speech resources, this approach has demonstrated improved recognition accuracy for the streaming ASR system when handling emotional utterances. Experiments conducted on both the MELD and SIMS datasets have shown a significant decrease in Word Error Rate (WER) and Character Error Rate(CER) when using the joint learning method compared to the optimized baseline. Specifically, the WER decreased by 1.27 on the MELD dataset and the CER by 0.58 on the SIMS dataset.