GitLab

- Put functional test logic into one place, `functional_impl.py`
- Tidy imports

* Fixed floor_divide deprecation warnings seen in pytest output

* Fixed warning in test_flanger_triangle_linear

The VAD function trims the input tensor to the first instance of voice
activity on any channel or item. Trimming batches this way may be
undesirable as the item with earliest activity will dominate. Either
way, the batch behaviour does not match the itemwise behaviour.

The VAD batch consistency tests currently pass out of coincidence, but
they specify incorrect behaviour. This commit removes them.

Use shorter input sequences to avoid time out error on CI

Merges lfilter and spectrogram classes together in the common implementation and modifies the cpu and gpu test definitions accordingly

1. `F.phase_vocoder` accepts Tensor with complex dtype.
    * The implementation path has been updated from #758 so that they share the same code path by internally converting the input Tensor to complex dtype and performing all the operation on top of it.
    * Adopted `torch.polar` for simpler Tensor generation from magnitude and angle.
2. Updated tests
    * librosa compatibility test for complex dtype and pseudo complex dtype
        * Extracted the output shape check test and moved it to functional so that it will be tested on all the combination of `{CPU | CUDA} x {complex64 | complex128}`
    * TorchScript compatibility test for `F.phase_vocoder` and `T.TimeStretch`.
    * batch consistency test for `T.TimeStretch`.

Co-authored-by: krishnakalyan3 <skalyan@cloudera.com>

This PR additionally adds batching to kaldi compliance resample interface.

* Refactor Kaldi compatibility tests
Co-authored-by: Jeff Hwang <jeffhwang@fb.com>

* Parameterize `test_sliding_window_cmn`

* Extract test naming function

* Pass a spectrogram to `F.sliding_window_cmn`

* Set manual seed for remaining rand calls in suite

Parameterize `test_create_fb` so each set of values are tested
independently. Also explicitly skip on older versions of librosa (< 0.7.2) when
`norm="slaney"`.

* Test with batches of differing items

Issues that occur when different items in a batch influence one another
will not present when a batch is composed of identical items. When
checking the consistency of batched behavior, in order to catch these
issues items should be different.

Thus, use different items for the `functional` batch consistency tests
wherever possible.

* Generate different white noise in each channel

Don't duplicate a single channel multiple times. Since this is used for
testing, generate different noise in each channel so data leakage
between channels can be detected.

* Parameterize batch size in batch consistency tests

Rather than creating a batch of 3 items in each test and slicing it to
test two different batch sizes at once, parameterize the batch size on
the TestFunctional class itself. This will generate a separate set of
tests for each batch size (better isolating failures) and removes a
leaky abstraction where the test calling `assert_batch_consistencies`
had to know to give it a batch size greater than 1.

* Check inputs too, to catch in-place operations

Check inputs to the batch consistency operations too, to ensure any
in-place operations operate the same on items as batches - not just that
they output the same result.

* Use much shorter sample for phaser test

Using a 5-second signal for the phaser test takes a long time on CPU,
much longer than the other batch consistency tests. Use a shorter signal
instead.

* Load dual-channel wav for VAD test

The stereo wav has two channels, slightly offset, so they'll count as
different items.

* Load wav using common_utils, not torchaudio.load

* Test pitch frequency with different freqs per item

The pitch frequency batch test was using the same frequency for each
item, which may not catch data leakage between items within a batch. Use
different frequencies so these kinds of issues would be triggered, just
like the other batch consistency tests.

* Explain justification for single-item batch

modified amplitude_to_DB to clamp per-item when a batch is provided
Co-authored-by: Vincent QB <vincentqb@users.noreply.github.com>

Also replaces `assert_allclose` with `assertEqual`.
Co-authored-by: krishnakalyan3 <skalyan@cloudera.com>

* Warn if create_fb_matrix produces a column whose weights are all zeros

See also https://github.com/librosa/librosa/issues/478

We have been running unit test with editable installation. (i.e. `python setup.py develop`), with which we missed issues like #842. 

This CC makes installation in CI non-editable, and change test directory structure so that the source code will not shadow the installed version of `torchaudio`. With simple `pytest test`, `pytest` modifies `sys.path` and prepend checked out repository, which shadows the installed version.

To remedy this, the whole test suites has been moved from `./test` to `./test/torchaudio_unittest`. This adds nice module structure to our test code and we can do absolute import in each test module, which makes it possible again to run test with `python -m unittest torchaudio_unittest/XXX.py`

This change does not affect the regular development process (`python setup.py develop` && `pytest test`)