L3DAS22 enhancement recipe

[neillu23]

No description provided.

[codecov]

Codecov Report

Merging #4269 (da2266f) into master (b757b89) will decrease coverage by 0.03%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master    #4269      +/-   ##
==========================================
- Coverage   82.80%   82.77%   -0.04%     
==========================================
  Files         463      474      +11     
  Lines       40608    40755     +147     
==========================================
+ Hits        33627    33733     +106     
- Misses       6981     7022      +41     

Flag	Coverage Δ
test_integration_espnet1	66.58% <ø> (-0.56%)	⬇️
test_integration_espnet2	49.30% <ø> (-2.34%)	⬇️
test_python	69.73% <ø> (+0.99%)	⬆️
test_utils	23.45% <ø> (-1.01%)	⬇️

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

Impacted Files	Coverage Δ
espnet2/enh/loss/criterions/abs_loss.py	80.00% <0.00%> (-7.50%)	⬇️
espnet2/bin/enh_scoring.py	88.50% <0.00%> (-5.87%)	⬇️
espnet2/fileio/rttm.py	32.55% <0.00%> (-4.40%)	⬇️
espnet2/samplers/abs_sampler.py	83.33% <0.00%> (-4.17%)	⬇️
espnet2/enh/separator/asteroid_models.py	28.94% <0.00%> (-3.56%)	⬇️
espnet2/text/phoneme_tokenizer.py	83.39% <0.00%> (-3.51%)	⬇️
espnet2/bin/asr_inference.py	84.83% <0.00%> (-3.43%)	⬇️
espnet/utils/training/iterators.py	43.75% <0.00%> (-3.31%)	⬇️
espnet2/train/preprocessor.py	36.71% <0.00%> (-2.56%)	⬇️
...2/tts/utils/parallel_wavegan_pretrained_vocoder.py	36.66% <0.00%> (-2.05%)	⬇️
... and 311 more

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

[Emrys365]

Thanks! I just made some minor comments.

[Emrys365]

Could you also add your recipe to egs2/README.md?

[ftshijt]

It is very good to have multiple configs, could you also add results (dev would be fine) in RESTULS.md?

[sw005320]

This recipe should have a special evaluation (scoring) script prepared by an organizer (or by making the exact same scoring scheme by ourselves).

[neillu23]

Thanks for the review and comments! I have addressed the comments and made two major changes to the PR:

1. Add the testing data from the L3DAS challenge.
2. Add metric.sh for calculating the challenge score.

[ftshijt]

Looks very cool! some minor suggestion

[neillu23]

Thanks for the comments! I have added the downloading data part in the recipe!

[sw005320]

@neillu23, some comments still exist, including the results.

[neillu23]

@neillu23, some comments still exist, including the results.

Thanks for reminding me! I haven't changed the data/temp to mktemp in local/l3das22_data_prep.sh because the files under the temp directory are used in the following script (local/l3das22_multi_channel.sh). Do you think we should pass the temp directory variable across the scripts?
I am training a fasnet and I will update the results soon.

[Emrys365]

LGTM in general. I think we can merge it after existing comments are addressed.

[Emrys365]

LGTM! I will merge it. Thanks a lot @neillu23