added en-ja to MUST_C_V2 dataset (ST) recipes

[jasonmusespresso]

What?

added and tested two new target languages

• updated dataset names
• create the vocab size for en-zh from 4k to 8k;

Why?

language expansion for must_c_v2

Attentional Enc-Dec (st_train_st_conformer_raw_en_ja_bpe_tc4000_sp)

• ST config: train_st_conformer.yaml
• Inference config: decode_st_conformer.yaml
• Download model and run inference:

cd espnet
git checkout c25f8762a6f9b7c7c5739fe3e1e72c077e566a60
pip install -e .
cd home/jbao/must_c_v2_st1
./run.sh --skip_data_prep false --skip_train true --download_model jasonmusespresso/must_c_v2_st_train_st_conformer_raw_en_ja_bpe_tc4000_sp_valid

• Environments
  • python version: 3.9.18 (main, Sep 11 2023, 13:41:44) [GCC 11.2.0]
  • espnet version: espnet 202402
  • pytorch version: pytorch 2.1.0

dataset	score	verbose_score
decode_st_conformer_st_model_valid.acc.best/tst-COMMON.en-ja	11.5	42.1/17.2/8.1/4.0 (BP = 0.932 ratio = 0.935 hyp_len = 51925 ref_len = 55563)
decode_st_conformer_st_model_valid.acc.best/tst-HE.en-ja	12.2	42.0/17.6/8.3/4.2 (BP = 0.960 ratio = 0.961 hyp_len = 12241 ref_len = 12744)

See also

[jasonmusespresso]

FYI @brianyan918

[weikeduo1]

must_c_v2/asr1 missing data.sh

[sw005320]

• @brianyan918, can you review it?

• @jasonmusespresso, please add the results and config files. Also, please upload the model and put the link to the results.

[sw005320]

See https://github.com/espnet/espnet/blob/master/egs2/aishell/asr1/README.md as a reference.

[sw005320]

Also, see https://github.com/espnet/espnet/blob/master/CONTRIBUTING.md#132-espnet2-recipes

[brianyan918]

LGTM thanks @jasonmusespresso

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 48.96%. Comparing base (f6f011d) to head (823898d).

Additional details and impacted files

@@             Coverage Diff             @@
##           master    #5648       +/-   ##
===========================================
+ Coverage   14.24%   48.96%   +34.72%     
===========================================
  Files         757      494      -263     
  Lines       69304    43593    -25711     
===========================================
+ Hits         9872    21346    +11474     
+ Misses      59432    22247    -37185     

Flag	Coverage Δ
test_integration_espnet2	48.96% <ø> (?)
test_python_espnetez	?

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

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[sw005320]

we should not change it with our local setup

[sw005320]

It looks good to me, but I want to have input from @brianyan918

@brianyan918,

• can you confirm the change for sacrebleu_opt_extra?
• compared with the other recipes, the README.md style is different, which is fine. However, I could not find the model link for all models. @jasonmusespresso seems to follow this style. Is it OK? I think it is better to provide the model link.

[brianyan918]

It looks good to me, but I want to have input from @brianyan918

@brianyan918,

• can you confirm the change for sacrebleu_opt_extra?
• compared with the other recipes, the README.md style is different, which is fine. However, I could not find the model link for all models. @jasonmusespresso seems to follow this style. Is it OK? I think it is better to provide the model link.

The sacrebleu_opt_extra looks good, but shouldn't there be some value passed from run.sh? @jasonmusespresso

[jasonmusespresso]

Details

@brianyan918 yes, added it. in my run, I pass it when calling ./run.sh (e.g. ./run.sh --stage 13 --stop-stage 17 --sacrebleu_opt_extra "-tok ja-mecab -l en-ja --smooth-method exp"...), which should also work.