Yiming Wang

> Is the objf the right sign now? Yes

> Also, check that the sum of the output of the network averages close to zero; certain bugs might cause it to be biased in a positive or negative direction....

> Whichever way you sum or average (reducing to a scalar), it should in principle be close to zero; but not necessarily for a given minibatch. > […](#) > On...

I just tested and didn't see the errors. Did you check if hydra-core>=1.0.7,

Good point! I may need some time to figure it out

I think Batchnorm is per dimension, so the masked part will not affect the unmasked part?

Oh OK. How does Kaldi deal with it? Guarantee same length within a batch so no padding?

@jedyang97 Makes sense and thanks for your suggestion! I will write a recipe with a small enough dataset, like TIMIT, once I have time (these days I am too busy:().

Sorry I might not have time until the end of this month. It would be great if someone else could do it during this time, or I can do it...

Thanks! Done. Yiming