Remove comparison dependencies and skipIfs

[samanklesaria]

As previous PRs have precomputed the results to various calls to external libraries, we no longer need to depend on those libraries when running the test suite. This PR removes instructions to install sox, librosa and kaldi in the linux test suite runner and removes the skipIf declarations on the associated external library comparison tests.

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/audio/4031

• 📄 Preview Python docs built from this PR

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

❌ 9 New Failures, 1 Unrelated Failure

As of commit c1049e9 with merge base 879faa0 ():

NEW FAILURES - The following jobs have failed:

• .github/workflows/bandit.yml (gh)
• .github/workflows/integration-test.yml (gh)
• .github/workflows/lint.yml (gh)
• .github/workflows/pr-labels.yml (gh)
• .github/workflows/unittest-macos-cpu.yml (gh)
• .github/workflows/unittest-windows-cpu.yml (gh)
• .github/workflows/unittest-windows-gpu.yml (gh)
• Build Aarch64 Linux Wheels / pytorch/audio / build-wheel-py3_9-cuda-aarch6412_9-aarch64 (gh)
  ImportError: /lib64/libm.so.6: version GLIBC_2.29' not found (required by /usr/local/cuda/lib64/libnvshmem_host.so.3)`
• Build documentation / build / Build doc (gh)
  At least one of the pre-conditions you specified did not hold

BROKEN TRUNK - The following job failed but were present on the merge base:

👉 Rebase onto the `viable/strict` branch to avoid these failures

• Build Aarch64 Linux Wheels / pytorch/audio / upload / upload-wheel-py3_9-cuda-aarch6412_9-aarch64 (gh) (trunk failure)
  Unable to download artifact(s): Artifact not found for name: pytorch_audio__3.9_cu129_aarch64

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

[samanklesaria]

We're now failing because we lack sklearn. But sklearn wasn't even in our dependencies before! All I can think of is that something else (like librosa) depended on it, and we were able to import it because pip installing librosa gave us access to sklearn too.

[NicolasHug]

On the sklearn issue, I hope this is the only once, but if that's the case then let's just ignore the corresponding test. We can probably just skip it and comment-out the offending import line in test/torchaudio_unittest/example/hubert/__init__.py ?

[samanklesaria]

We can't just remove the change to the python path, as that will break test/example/hubert/test_crop_audio_label.py. But the test itself doesn't use the kmeans functionality. The learn_kmeans function that uses sklearn is never tested. So as long as we hide the import inside the call to learn_kmeans, we're okay.

[samanklesaria]

It seems like we need to remove the backends. If they're in the repository, they'll get added to the deprecations tests. But because the dependencies are there, we'll get errors. But that's a much larger change than we probably want to include in this PR. If we want to make the CI green for this PR without removing the sox backend from the codebase, the easiest fix I see is a hack to make the deprecation tests ignore anything from a sox module.

[samanklesaria]

It also seems like the kaldi sliding window test expected results we're properly added when we merged. This is probably because in the upstream commit (the librosa tests) I changed the naming convention to remove cpu/cuda naming differences. And the kaldi files were't regenerated after the change.

[samanklesaria]

Verbally approved by Nicholas on slack. Merging.