Speaker_Task, domain_base and emotion_prediction

[SuFame920]

Hi author, thanks for your contribution to ERC. Recently, I tried to reproduce the results of your paper, but encountered the following problems:

1. with all three parameters mentioned above being False, LoRa based on LLaMA-2 can achieve an average accuracy of 67（iemocap）, which proves that the authors' work has a great potential, since speaker recognition, related retrieval, and emotional tendency prediction tasks have not been introduced yet;
2. but the above three parameters, whenever set to true, even a single setting of one, will report an error in data_process.py, as shown below
3. In addition, I noticed that data_process.py about the processing of the above three tasks are not specifically written out, are pass, here whether we need to be based on the thesis idea of their own realization before using the above three modules?