while running run.sh in gop_speechocean it get error in visualize_feat.py AttributeError: 'tuple' object has no attribute 'shape' about gopt

[YuanGongND]

If it occurs in gop_speechocean762/s5$ ./run.sh, then it should be a question to Kaldi maintainer. Are you test with your own data?

[amandeepbaberwal]

No i am testing with speechocean, using 1 train and 1 test data

…

Sent from my iPhone On 19-Mar-2023, at 1:04 AM, Yuan Gong ***@***.***> wrote:  If it occurs in gop_speechocean762/s5$ ./run.sh, then it should be a question to Kaldi maintainer. Are you test with your own data? — Reply to this email directly, view it on GitHub<#20 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/ACGTJQGRBETAO7E6AJXHCYTW4YE4LANCNFSM6AAAAAAV7PJNIA>. You are receiving this because you authored the thread.Message ID: ***@***.***>

[YuanGongND]

See the first line of warning utils/validate_data_dir.sh: WARNING: you have only one speaker. This probably a bad idea., it might due to you only use part of the data. Using the entire dataset might be a better idea since it is not large.

But again, this should be a question to https://github.com/kaldi-asr/kaldi/issues?q=is%3Aissue+gop , they might updated their code and I am not part of the recipe developers.

[infinite-darkness108]

solved! In local/visualize_feats.py, add features=np.array(features)
You should now be able to see results in

1. exp/gop_train/feats.png (t-SNE plot for two phones; the higher quality ones (as scored by 5 human experts) being well separated and low scored ones being in the mix of two clusters.)
2. exp/gop_test/result_gop.txt (It is in the format

<utt_id>.0 2.0 1.0
<utt_id>.1 2.0 2.0
.
.
<utt_id>.n 1.0 0.0
.
.
.
so on
)
meaning, the utterance's zeroth phone was scored 2.0 by human expert (avg of 5) and the prediction by the GoP based approach is 1.0, and the same utterance's next canonical phone was given a score of 2.0 by both human experts and the model.