Skip to content

Conversation

@hainan-xv
Copy link
Contributor

This was first tested on the material recipe, and also on swbd. It was giving the almost identical performance as before but faster.

Copy link
Contributor

@danpovey danpovey left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

thanks. A couple small comments.

int32 lm_state;

// the distance between the state from the start state
int32 depth;
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

it might be more accurate to describe this as:

the number of arcs on the path from the start state to this state, in the composed lattice, by which this state was first reached. 
// in a depth-first way. Once we find a path reaching a final state, this
// variable will be reset to 0.
// The reason we do this is because the beam-search depends on a good estimate
// of the composed-best-cost, which before we reach a fina state, we instead
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

final

@danpovey danpovey merged commit 9210f3a into kaldi-asr:master Jun 15, 2018
@hainan-xv hainan-xv deleted the new-composition branch June 17, 2018 00:21
dpriver pushed a commit to dpriver/kaldi that referenced this pull request Sep 13, 2018
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

2 participants