speech-emotion-recognition-using-self-attention
speech-emotion-recognition-using-self-attention copied to clipboard
KrishnaDN
the result about the code
hi, I cannot get the same results as the paper,too. And there is a overfitting. As the train WA is over 80%, but the test WA is about 50%. Do you have the same proplem? finaly, I want ask your result about the WA and UA?
hi, I cannot get the same results as the paper,too. And there is a overfitting. As the train WA is over 80%, but the test WA is about 50%. Do you have the same proplem? finaly, I want ask your result about the WA and UA?
Hi Jingyu, I have also faced the same problem and I have wrote a mail to the authors about it. I even asked them for the code but due to patent work they said they can't give us their codes.
ok, how about the results without Multitask learning in this paper. Have you got the WA about 70.5% ?
hi, I cannot get the same results as the paper,too. And there is a overfitting. As the train WA is over 80%, but the test WA is about 50%. Do you have the same proplem? finaly, I want ask your result about the WA and UA?
Hi Jingyu, I have also faced the same problem and I have wrote a mail to the authors about it. I even asked them for the code but due to patent work they said they can't give us their codes.
ok, how about the results without Multitask learning in this paper. Have you got the WA about 70.5% ?
Without multitask you should get about 51% UA according to the paper. Their claim is that, the multi-task component is giving huge improvement. As per the authors mail, that is what they are saying. So , stand alone self attention is not a great add on here.
Thank you for the relpy, krishnaDN. But the fact is the results didn't got a big improvemen even adding the multi-task learning.
Thank you for the relpy, krishnaDN. But the fact is the results didn't got a big improvemen even adding the multi-task learning.
That's what I am also surprised about. I tried asking them their code but due to various reasons they can't provide. I think you should talk to the authors once, things may become more clear
Thank you for the relpy, krishnaDN. But the fact is the results didn't got a big improvemen even adding the multi-task learning.
That's what I am also surprised about. I tried asking them their code but due to various reasons they can't provide. I think you should talk to the authors once, things may become more clear
ok, I will write to the author. Hope we can find the answer.
Hi,@jingyu-95 ,did you get any answer or make progress? Could you please share your complete codes? Thanks.
