Closed xfwu closed 8 years ago
I just checked in new updates to do so. The high-level script is align_ctc_single_utt.sh Note that currently we can only generate alignments utterance by utterance, rather than in a batch mode (as in Kaldi)
Thank you very much!
I presume that the uttdata-dir stores only one utterance's reference, therefore the G.fst generated out of this reference will be used to generate only a very compact lattice, and the the 1best path found in that lattice is actually the reference!!
this is very cool!
Best
Yep, your understanding is correct.
What is the best way to do a force alignment/decoding in Eesen? I want to generate the alignment of the reference, given the softmaxed output
Best