Re: [kaldi-help] a query about command lattice-to-ctm-conf

[Daniel Povey]

If that's what you want, it would be more convenient to deal with n-best lists.  See lattice-to-nbest.

On Mon, Apr 17, 2017 at 8:41 PM, <ntu.spe...@gmail.com> wrote:

Dear all, 

lattice-to-ctm-conf --acoustic-scale=0.1 --decode-mbr=false  ark:1.lats ark:1.1best 1.ctm, 

The CTM file contains word or phone-level posteriors for the units in 1.lats. 

My question is: the 1.best file is achieved from decoding? shall we produce artificial 1.best file, For example, assuming the 1-best phone sequence is "big apple", after executing the above commands,  posterior is produced. Now, I want to know the  second best phone sequence's posterior, e.g. "big aple". is it possible to realize this?

Many thanks

--
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+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[ISDPL]

Dear Povey, 

Recently, I am also playing with lattice-to-ctm-conf, it provides two usages: with/without 1best-rspecifier. Without 1best-rspecifier, this command use Minimum Bayes Risk decoding and get best path

with corresponding scores. Otherwise, this command will use provided 1best-rspecifier to calculate scores.

My question is file 1best-rspecifier produced by command lattice-1best?

Thanks

在 2017年4月18日星期二 UTC-4下午12:23:25，Dan Povey写道：

To unsubscribe from this group and stop receiving emails from it, send an email to kaldi-help+...@googlegroups.com.

[Daniel Povey]

I forgot about that option (or maybe never knew about it).  Yes, that is the same format.