On the performance of speaker recognition tasks in MELD

[SuFame920]

Many thanks to the authors and to your team for their outstanding contributions to the ERC field! We are happy that we roughly reproduced the experimental results, although there is roughly 1-2% difference, but this was done without parameterization and the results are still exciting. 
However, we found that the speaker recognition task performs poorly on MELD's multi-speaker recognition,even affecting the accuracy of the model, as shown in the results below. ME. Therefore, when the speaker recognition task is removed, the model accuracy is slightly improved instead, which is different from what the authors reported in their paper, perhaps because of our different experimental settings, and we hope the authors can solve the doubts.