🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/audio/4037

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

❌ 2 New Failures, 1 Pending, 8 Unrelated Failures

As of commit ec301d8 with merge base bc3de07 ():

This comment was automatically generated by Dr. CI and updates every 15 minutes.

We have to merge #4031 first.

It seems that CTC Decoder requires flashlight-text package, which it depends on pretty heavily. We could disable the tests for it.

test_librispeech_lightning and test_mustc_lightning and test_tedlium3_lightning require lightning. It seems these just test lightning integration (which we're deprecating anyway), so we should be able to delete these tests.

tacotron2/test_text_preprocessing.py requires unidecode. But we're only testing on ascii text. Would could give text_to_sequence a flag for whether or not to use unidecode? Or we could just add a warning to it saying that only ascii input is allowed?

I'm still not sure how much we're testing the use of inflect, which is also used in tacotron2/test_text_preprocessing.py. Perhaps we can restrict our testing to examples that don't require it, and not load the module until it is needed.

OK, thanks. Let's just skip all of those by setting the appropriate env variable e.g. TORCHAUDIO_TEST_ALLOW_SKIP_IF_NO_CTC_DECODER=1.