Add ASR recipe of MAGICDATA mandarin read speech

[tjysdsg]

http://www.openslr.org/68/

Most of the code is copied from aishell2 recipe

TODOs:

• Store training results in README.md
  • Preliminary results
  • Fix training loss "hill climb"
• Automatically download data and unzip

[codecov]

Codecov Report

Merging #4635 (17f568c) into master (ef6c15f) will increase coverage by 0.00%.
The diff coverage is n/a.

@@           Coverage Diff           @@
##           master    #4635   +/-   ##
=======================================
  Coverage   83.09%   83.10%           
=======================================
  Files         518      518           
  Lines       44700    44700           
=======================================
+ Hits        37145    37146    +1     
+ Misses       7555     7554    -1     

Flag	Coverage Δ
test_integration_espnet1	66.37% <ø> (ø)
test_integration_espnet2	49.36% <ø> (ø)
test_python	70.92% <ø> (+<0.01%)	⬆️
test_utils	23.30% <ø> (ø)

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

Impacted Files	Coverage Δ
espnet/distributed/pytorch_backend/launch.py	83.90% <0.00%> (+1.14%)	⬆️

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[tjysdsg]

@ftshijt Hi Jiatong, thanks for the review. I'll fix these issues in future commits and train a model using the new configuration

[ftshijt]

If you think the PR is ready, you can remove the WIP tag in the title. And I can proceed to a final review.

[tjysdsg]

If you think the PR is ready, you can remove the WIP tag in the title. And I can proceed to a final review.

The PR is not ready yet, since the training loss of the second model I trained yesterday goes up significantly after 7epoch. The 7-th epoch does seem to show better performance than the first model I trained.

Now the cluster is back online, I'm training another one with a new learning rate and warmup_step. I'll let you know if the PR is ready for review. Thanks

[tjysdsg]

@ftshijt Hi Jiatong, this PR is now ready for review. Thanks!

[ftshijt]

Many thanks for the contribution. LGTM! I will merge it once the CI is passed

[tjysdsg]

@ftshijt Seems like the CI script itself could be broken? All recent commits to the master branch failed to pass CI.

[ftshijt]

Yeah, understand, we will have people working on fixing the CI. Please allow some time for it to be fixed.

[tjysdsg]

@ftshijt Hi, the model has been uploaded to HuggingFace. No further PR is needed :)

[ftshijt]

Everything seems to be good on my side. Will be wait the CI to be fixed

[ftshijt]

Can you merge the latest Commit to fix the CI issues?

[tjysdsg]

@ftshijt Hi, sorry for the late reply and thanks for merging it for me!