Add streaming Transformer model using tedlium2

[codecov-commenter]

Codecov Report

Merging #57 (01a6cfd) into master (7e956cf) will not change coverage. The diff coverage is n/a.

@@           Coverage Diff           @@
##           master      #57   +/-   ##
=======================================
  Coverage   51.04%   51.04%           
=======================================
  Files           2        2           
  Lines         384      384           
=======================================
  Hits          196      196           
  Misses        188      188           

---

Continue to review full report at Codecov.

Legend - Click here to learn more Δ = absolute <relative> (impact), ø = not affected, ? = missing data Powered by Codecov. Last update 7e956cf...01a6cfd. Read the comment docs.

[sw005320]

Please fix https://github.com/espnet/espnet_model_zoo/runs/4211725385?check_suite_focus=true#step:8:66

[D-Keqi]

Please fix https://github.com/espnet/espnet_model_zoo/runs/4211725385?check_suite_focus=true#step:8:66

Sure! But I still do not know why I got a CI issue. I just append a line to the table.csv and I have tested the CI locally, in which no issue was reported.

[sw005320]

Please fix https://github.com/espnet/espnet_model_zoo/runs/4211725385?check_suite_focus=true#step:8:66

Sure! But I still do not know why I got a CI issue. I just append a line to the table.csv and I have tested the CI locally, in which no issue was reported.

The error message says that the model you added does not follow the correct format probably due to that the model was uploaded at huggingface hub. @Fhrozen, can you help us to fix it? The test does not seem to consider the model in huggingface hub.

[D-Keqi]

Please fix https://github.com/espnet/espnet_model_zoo/runs/4211725385?check_suite_focus=true#step:8:66

Sure! But I still do not know why I got a CI issue. I just append a line to the table.csv and I have tested the CI locally, in which no issue was reported.

The error message says that the model you added does not follow the correct format probably due to that the model was uploaded at huggingface hub. @Fhrozen, can you help us to fix it? The test does not seem to consider the model in huggingface hub.

I see. I used to upload the model via zenodo, but It seems that there are some issues with zenodo. So I chose the huggingface as encouraged by the documents.

[Fhrozen]

The problem is with the use of the class for pretrained models on espnet (i.e. Speech2Text). Until now the downloader works well and for our experiments we did not use that description for test. https://github.com/espnet/espnet/blob/master/espnet2/main_funcs/pack_funcs.py#L18-L36

It will be required to add that in case of a directory (such as the one downloaded from huggingface) skip any extraction: https://github.com/espnet/espnet/blob/713e784c0815ebba2053131307db5f00af5159ea/espnet2/main_funcs/pack_funcs.py#L55-L61

[sw005320]

@Fhrozen, can you fix this? Can we change the test for the HF hub case?

[Fhrozen]

sure, I will submit a PR for this issue.

[sw005320]

It still has an issue https://github.com/espnet/espnet_model_zoo/runs/4394917703?check_suite_focus=true#step:8:95

[D-Keqi]

It still has an issue https://github.com/espnet/espnet_model_zoo/runs/4394917703?check_suite_focus=true#step:8:95

I have fixed it again. Sorry, can you approve it again for a test? @Fhrozen

[D-Keqi]

I found the issue. I need to change the tag of the streaming Transformer model uploaded by Emiru to fix the CI issue.

[sw005320]

Thanks, @D-Keqi and @Fhrozen!