[WIP] Iam_ocr Adding Wellington corpus #2354
Conversation
merging the files from the IAM_eng branch to use that file structure
…keep consistent with aarora8/IAM_eng branch
Keeping up to date with code written by aarora8
…le-diag-gmm.h instead of gmm-est.cc
…amed run.sh to run{1-6}.sh
…ting instead of lines of text
… the LOB corpus entirely
egs/iam/v1/run.sh
Outdated
| steps/decode.sh --nj $nj --cmd $cmd exp/mono/graph data/test \ | ||
| exp/mono/decode_test | ||
| fi | ||
| #if [ $stage -le 5 ]; then |
There was a problem hiding this comment.
are these commented-out parts supposed to be commented out?
There was a problem hiding this comment.
Yes, I intentionally commented out the decoding parts because the results from the first couple of stages (monophone/triphone) are not very good and I didn't think I needed to compute them. I can just pass their alignments on to the later CNN stages of the script.
There was a problem hiding this comment.
Rather than comment them out, you can set at the top
decode_gmm=false
and do things like
if [ $stage -le 5 ] && $decode_gmm; then
| @@ -0,0 +1,138 @@ | |||
| #!/usr/bin/env python3 | |||
There was a problem hiding this comment.
A cleaned version of this file './local/remove_test_utterances_from_lob.py' is already part of kaldi/master : kaldi/egs/iam/v1.
|
I am thinking of merging this within 24 hours-- I want to be a bit more aggressive about merging these recipes as I think it will help our work move faster. |
|
oh wait, I see this PR is very small. Merging. |
4 participants
I haven't updated the new results on the top of the scripts yet.