Add Libriheavy small and medium ASR2 recipes

[akreal]

What?

ASR2 recipe for Libriheavy medium subset.

Why?

I'm running experiments with a causal LM and ASR2 on this dataset and would like to have a comparison point without a causal LM.

See also

https://github.com/k2-fsa/libriheavy

I'll add results and a model link in about a week.

[codecov]

Codecov Report

Merging #5512 (10c3bb8) into master (0d0ab98) will decrease coverage by 12.34%.
Report is 5 commits behind head on master.
The diff coverage is n/a.

@@             Coverage Diff             @@
##           master    #5512       +/-   ##
===========================================
- Coverage   70.31%   57.97%   -12.34%     
===========================================
  Files         711      710        -1     
  Lines       65757    65670       -87     
===========================================
- Hits        46237    38074     -8163     
- Misses      19520    27596     +8076     

Flag	Coverage Δ
test_integration_espnet2	48.61% <ø> (ø)
test_python_espnet1	?
test_python_espnet2	51.36% <ø> (+<0.01%)	⬆️
test_utils	22.19% <ø> (ø)

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

see 130 files with indirect coverage changes

📣 Codecov offers a browser extension for seamless coverage viewing on GitHub. Try it in Chrome or Firefox today!

[akreal]

I also added the recipe for the small subset. It should be ready for review now.

The official baseline seems to do case sensitive scoring for the non-normalized text, so I added -s to score_opts in these recipes.

sclite skips all lines starting with ** and prints a warning for all lines starting with * (it's hardcoded as a comment character). I've added the removal of * in beginnings of texts in the scoring stage of asr2.sh. I've manually verified that it works as intended on the dev set, it should have no effect on other recipes.

[sw005320]

Thanks, @akreal!