GOP recipe issues of compute-gop

20 views
Skip to first unread message

pavitra kulkarni

unread,
Jun 29, 2025, 9:07:44 AMJun 29
to kaldi...@googlegroups.com
Hi All,
I am trying to implement GOP recipe.In the compute-gop stage (stage 12)I am getting below issues:

nnet3-compute --frame-subsampling-factor=3 \
    exp/chain_cleaned/tdnn_1d_sp/final.raw \
    scp:exp/feats_train/raw_feats.1.scp ark:- | \
compute-gop --phone-map=data/lang_nosp/phone-to-pure-phone.int \
    --skip-phones-string=0:1:2 \
    exp/chain_cleaned/tdnn_1d_sp/final.mdl \
    "ark,t:gunzip -c exp/ali_train/ali.1.gz|" \
    "ark,t:gunzip -c exp/ali_train/ali-phone.1.gz|" \
    ark:- \
    ark,scp:exp/gop_train/gop.1.ark,exp/gop_train/gop.1.scp \
    ark,scp:exp/gop_train/feat.1.ark,exp/gop_train/feat.1.scp

ERROR (nnet3-compute[5.5.1058~1-aa17]:EnsureFrameIsComputed():nnet-am-decodable-simple.cc:105) Neural net expects 'ivector' features with dimension 100 but you provided 0

Then I give,
nnet3-compute --frame-subsampling-factor=3 \
    --online-ivector=scp:exp/feats_train/ivector.1.scp \
    exp/chain_cleaned/tdnn_1d_sp/final.raw \
    scp:exp/feats_train/raw_feats.1.scp ark:- | \

ERROR (nnet3-compute[5.5.1058~1-aa17]:Read():parse-options.cc:366) Invalid option --online-ivector=scp:/home/user/Models_208/GOP/data/train/ivectors/ivector_online.JOB.scp

nnet3-compute shows --ivectors,--online-ivector options,both gives the same issue.

I also tried pasting both feats and ivectors as below

bash -c "paste-feats --length-tolerance=100000         scp:exp/feats_train/raw_feats.1.scp         scp:exp/feats_train/ivector.1.scp ark:- |       nnet3-compute --frame-subsampling-factor=3         /home/user/Models_208/GOP/database/exp/chain_cleaned/tdnn_1d_sp/final.raw ark:- ark:- |       compute-gop --phone-map=data/lang_nosp/phone-to-pure-phone.int         --skip-phones-string=0:1:2         /home/user/Models_208/GOP/database/exp/chain_cleaned/tdnn_1d_sp/final.mdl         'ark,t:gunzip -c exp/ali_train/ali.1.gz|'         'ark,t:gunzip -c exp/ali_train/ali-phone.1.gz|'         ark:-         ark,scp:exp/gop_train/gop.1.ark,exp/gop_train/gop.1.scp         ark,scp:exp/gop_train/feat.1.ark,exp/gop_train/feat.1.scp"

Now the error is:
ERROR (nnet3-compute[5.5]:EnsureFrameIsComputed():nnet-am-decodable-simple.cc:101) Neural net expects 'input' features with dimension 40 but you provided 140

How to resolve this,Kindly let me know.


--
Regards
pavitra kulkarni
Reply all
Reply to author
Forward
0 new messages