Update stft tests to support latest librosa (#72833)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/72833

Closes #72550

The latest version of librosa breaks backward compatibility in two
ways:
- Everything except the input tensor is now keyword-only
- `pad_mode` now defaults to `'constant'` for zero-padding

https://librosa.org/doc/latest/generated/librosa.stft.html

This changes the test to match the old behaior even when using the new
library and updates the documentation to explicitly say that
`torch.stft` doesn't exactly follow the librosa API. This was always
true (`torch.stft` it has new arguments, a different default window
and supports complex input), but it can't hurt to be explicit.

Test Plan: Imported from OSS

Reviewed By: ngimel

Differential Revision: D34386897

Pulled By: mruberry

fbshipit-source-id: 6adc23f48fcb368dacf70602e9197726d6b7e0c1
(cherry picked from commit b5c5ed41963022c9f26467279ed098fb905fa00a)

test/test_spectral_ops.py

@@ -879,9 +879,16 @@ class TestFFT(TestCase):
             input_1d = x.dim() == 1
             if input_1d:
                 x = x.view(1, -1)
+
+            # NOTE: librosa 0.9 changed default pad_mode to 'constant' (zero padding)
+            # however, we use the pre-0.9 default ('reflect')
+            pad_mode = 'reflect'
+
             result = []
             for xi in x:
-                ri = librosa.stft(xi.cpu().numpy(), n_fft, hop_length, win_length, window, center=center)
+                ri = librosa.stft(xi.cpu().numpy(), n_fft=n_fft, hop_length=hop_length,
+                                  win_length=win_length, window=window, center=center,
+                                  pad_mode=pad_mode)
                 result.append(torch.from_numpy(np.stack([ri.real, ri.imag], -1)))
             result = torch.stack(result, 0)
             if input_1d:

torch/functional.py

@@ -586,7 +586,8 @@ def stft(input: Tensor, n_fft: int, hop_length: Optional[int] = None,
 
     The STFT computes the Fourier transform of short overlapping windows of the
     input. This giving frequency components of the signal as they change over
-    time. The interface of this function is modeled after the librosa_ stft function.
+    time. The interface of this function is modeled after (but *not* a drop-in
+    replacement for) librosa_ stft function.
 
     .. _librosa: https://librosa.org/doc/latest/generated/librosa.stft.html