How to train the model?

[Guodong0712]

Hi
Thank you in advance,
I am trying to implement the X-vector using your TDNN model, but when I try to train it and the loss can not down and the rate of right is so low.
Thank you, could you give me some suggestions?
my input: 24-dim MFCC

                                                                                                                                    Yours sincerely
                                                                                                                                             Ma

[qinxiaoyi]

Hi
Thank you in advance,
I am trying to implement the X-vector using your TDNN model, but when I try to train it and the loss can not down and the rate of right is so low.
Thank you, could you give me some suggestions?
my input: 24-dim MFCC

                                                                                                                                    Yours sincerely
                                                                                                                                             Ma

i have the same question, when i train the model, the loss always nan, than i exchange the relu and batchnorm, but the eer is very high

[Guodong0712]

Thanks for your reply, I am a freshman and have not to compute the eer because the final exam.

[Guodong0712]

Hi
Thank you in advance,
I am trying to implement the X-vector using your TDNN model, but when I try to train it and the loss can not down and the rate of right is so low.
Thank you, could you give me some suggestions?
my input: 24-dim MFCC

                                                                                                                                    Yours sincerely
                                                                                                                                             Ma

i have the same question, when i train the model, the loss always nan, than i exchange the relu and batchnorm, but the eer is very high

Hi，thank you first, I make the score with cosine, and about 0.998 between the same speaker utt and
0.993 between the different, so I predict that eer will be high, can you give me some suggestions?

Ma

[HappyGallery]

Hi
Thank you in advance,
I am trying to implement the X-vector using your TDNN model, but when I try to train it and the loss can not down and the rate of right is so low.
Thank you, could you give me some suggestions?
my input: 24-dim MFCC

                                                                                                                                    Yours sincerely
                                                                                                                                             Ma

i have the same question, when i train the model, the loss always nan, than i exchange the relu and batchnorm, but the eer is very high

Hi ，if you have solved the question ,loss always nan .I have meet the problem too , but i have no method to solve . if you know how please teach me . Thanks!

[HappyGallery]

@qinxiaoyi