Migrate kaldi spectrogram

[bhargavkathivarapu]

This PR migrates the kaldi spectrogram test (#597 ) from test/test_compliance_kaldi.py to test/kaldi_compatibility_impl.py

[codecov]

Codecov Report

Merging #687 into master will increase coverage by 0.00%.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##           master     #687   +/-   ##
=======================================
  Coverage   88.79%   88.80%           
=======================================
  Files          22       22           
  Lines        2356     2358    +2     
=======================================
+ Hits         2092     2094    +2     
  Misses        264      264           

Impacted Files	Coverage Δ
torchaudio/compliance/kaldi.py	96.09% <100.00%> (+0.02%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update b4284de...2725ba9. Read the comment docs.

[bhargavkathivarapu]

@mthrok , Applied device and dtype patch to kaldi spectrogram function in kaldi.py . Removed the failing test cases .

[mthrok]

Mostly looks good.

[mthrok]

This change does not seem to be necessary.
strided_input and fft should be already on the right dtype/device.
I tried your code without this change and tests ran fine.
Did you find a case this causes RuntimeError?

[bhargavkathivarapu]

Reverted this change,unittest passed.
Before I got runtime error at torch.max step , so safe side I changed device and dtype for inputs to max

[mthrok]

I tried running it with CUDA and test_spectrogram_015 and test_spectrogram_107 are failing. Can you remove them too?

$pytest -v test/kaldi_compatibility_cuda_test.py -k spec

_________________ TestKaldi_CUDA_Float32.test_spectrogram_015 __________________

a = (<test.common_utils.TestKaldi_CUDA_Float32 testMethod=test_spectrogram_015>,)

    @wraps(func)
    def standalone_func(*a):
>       return func(*(a + p.args), **p.kwargs)

/home/moto/conda/envs/PY3.8-cuda101/lib/python3.8/site-packages/parameterized/parameterized.py:530:
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
test/kaldi_compatibility_impl.py:98: in test_spectrogram
    self.assert_equal(result, expected=kaldi_result, rtol=1e-4, atol=1e-8)
test/kaldi_compatibility_impl.py:60: in assert_equal
    self.assertEqual(output, expected, rtol=rtol, atol=atol)
../pytorch/torch/testing/_internal/common_utils.py:1083: in assertEqual
    self.assertTrue(result, msg=message)
E   AssertionError: False is not true : Tensors failed to compare as equal! With rtol=0.0001 and atol=1e-08, found 2 element(s) (out of 18) whose difference(s) exceeded the margin of error (including 0 nan comparisons). The greatest difference was 2.384185791015625e-05 (0.20874881744384766 vs. 0.2087249755859375), which occurred at index (1, 5).
----------------------------- Captured stderr call -----------------------------
compute-spectrogram-feats --blackman-coeff=0.7797 --dither=0 --energy-floor=2.5119 --frame-length=0.625 --frame-shift=0.5 --preemphasis-coefficient=0.69 --raw-energy=true --remove-dc-offset=false --round-to-power-of-two=false --snip-edges=false --subtract-mean=true --window-type=povey scp:- ark:-
LOG (compute-spectrogram-feats[5.5.689~1-2c7e7]:main():compute-spectrogram-feats.cc:157)  Done 1 out of 1 utterances.
_________________ TestKaldi_CUDA_Float64.test_spectrogram_107 __________________

a = (<test.common_utils.TestKaldi_CUDA_Float64 testMethod=test_spectrogram_107>,)

    @wraps(func)
    def standalone_func(*a):
>       return func(*(a + p.args), **p.kwargs)

/home/moto/conda/envs/PY3.8-cuda101/lib/python3.8/site-packages/parameterized/parameterized.py:530:
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
test/kaldi_compatibility_impl.py:98: in test_spectrogram
    self.assert_equal(result, expected=kaldi_result, rtol=1e-4, atol=1e-8)
test/kaldi_compatibility_impl.py:60: in assert_equal
    self.assertEqual(output, expected, rtol=rtol, atol=atol)
../pytorch/torch/testing/_internal/common_utils.py:1083: in assertEqual
    self.assertTrue(result, msg=message)
E   AssertionError: False is not true : Tensors failed to compare as equal! With rtol=0.0001 and atol=1e-08, found 1 element(s) (out of 28) whose difference(s) exceeded the margin of error (including 0 nan comparisons). The greatest difference was 1.1195481270931396e-06 (0.008482767170622907 vs. 0.00848388671875), which occurred at index (4, 0).
----------------------------- Captured stderr call -----------------------------
compute-spectrogram-feats --blackman-coeff=4.5882 --dither=0 --energy-floor=1.0111 --frame-length=0.375 --frame-shift=0.1875 --preemphasis-coefficient=0.78 --raw-energy=false --remove-dc-offset=true --round-to-power-of-two=false --snip-edges=false --subtract-mean=true --window-type=blackman scp:- ark:-
LOG (compute-spectrogram-feats[5.5.689~1-2c7e7]:main():compute-spectrogram-feats.cc:157)  Done 1 out of 1 utterances.
=========================== short test summary info ============================
FAILED test/kaldi_compatibility_cuda_test.py::TestKaldi_CUDA_Float32::test_spectrogram_015
FAILED test/kaldi_compatibility_cuda_test.py::TestKaldi_CUDA_Float64::test_spectrogram_107
================ 2 failed, 220 passed, 178 deselected in 13.96s ================

[bhargavkathivarapu]

Removed those 2 tests and reverted the device change for fft step

[mthrok]

Removed those 2 tests and reverted the device change for fft step

Looks good and verified all tests pass on CUDA. Thanks!