[WIP] Add Aishell-2 ASR Recipe for Espnet2

[jctian98]

Hi, dear Espnet team.
I'm trying to make a PR for Espent2 Aishell-2 recipe.
link for the dataset: https://www.aishelltech.com/aishell_2

[codecov]

Codecov Report

Merging #4451 (40c5f69) into master (09b6f64) will increase coverage by 0.09%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master    #4451      +/-   ##
==========================================
+ Coverage   82.67%   82.77%   +0.09%     
==========================================
  Files         474      474              
  Lines       40755    40755              
==========================================
+ Hits        33693    33733      +40     
+ Misses       7062     7022      -40     

Flag	Coverage Δ
test_integration_espnet1	66.58% <ø> (ø)
test_integration_espnet2	49.30% <ø> (ø)
test_python	69.73% <ø> (+0.50%)	⬆️
test_utils	23.45% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
espnet2/enh/layers/beamformer.py	84.09% <0.00%> (+0.32%)	⬆️
espnet/nets/pytorch_backend/e2e_asr_transformer.py	96.08% <0.00%> (+0.35%)	⬆️
espnet2/diar/espnet_model.py	96.29% <0.00%> (+0.61%)	⬆️
espnet2/gan_tts/vits/vits.py	100.00% <0.00%> (+0.77%)	⬆️
espnet2/enh/layers/complex_utils.py	80.73% <0.00%> (+0.91%)	⬆️
espnet2/layers/stft.py	73.73% <0.00%> (+1.01%)	⬆️
espnet2/hubert/espnet_model.py	93.13% <0.00%> (+1.96%)	⬆️
.../nets/pytorch_backend/transformer/encoder_layer.py	98.07% <0.00%> (+3.84%)	⬆️
espnet/nets/pytorch_backend/e2e_asr_maskctc.py	91.85% <0.00%> (+4.44%)	⬆️
espnet2/bin/asr_inference_k2.py	93.07% <0.00%> (+4.61%)	⬆️
... and 1 more

📣 Codecov can now indicate which changes are the most critical in Pull Requests. Learn more

[sw005320]

LGTM.
@ftshijt, can you take a look at this?

Once it is ready, we may merge this PR and have a follow-up PR to update a result and upload a model.

[jctian98]

Hi! I have uploaded the README file and the model files for the hybrid CTC-Attention model. The results and model files of the transducer model would be updated in days.

[ftshijt]

Thanks for your contribution! I have some minor issues and suggestions as follows:

[jctian98]

All fixed. Thanks :)

[ftshijt]

Everything looks find for me. After the CI checks, I will merge it for now as we already got some initial results. You can start another follow-up PRs for other models and results. Many thanks!

[sw005320]

Thanks, @jctian98!