--
Go to http://kaldi-asr.org/forums.html find out how to join
---
You received this message because you are subscribed to the Google Groups "kaldi-help" group.
To unsubscribe from this group and stop receiving emails from it, send an email to kaldi-help+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/kaldi-help/317d19ac-8c2d-4512-af4e-3c72308bd4d0%40googlegroups.com.
t is something inbuilt to the framework, it's the time index. It means "get the ivector from time t=0, regardless of the t value that we're asked to provide at the output of this expression."The affine transform thingy is something similar to LDA, but not quite the same. You don't need to worry about it; recent experiments by Desh (cc'd) show that we can get as good or better performance by computing conventional delta and delta-delta features, which is more standard.
Dan
On Mon, Jul 22, 2019 at 7:39 AM Anjos <aqh...@gmail.com> wrote:
--fixed-affine-layer name=lda input=Append(-2,-1,0,1,2,ReplaceIndex(ivector, t, 0)) affine-transform-file=exp/nnet3/tdnn_sp/configs/lda.matReplaceIndex(ivector, t, 0) ,the variable t is what?the affine-transform is how to do?
Go to http://kaldi-asr.org/forums.html find out how to join
---
You received this message because you are subscribed to the Google Groups "kaldi-help" group.
To unsubscribe from this group and stop receiving emails from it, send an email to kaldi...@googlegroups.com.