Move Seed to Setup (#2425)

Summary:
Bringing in move seed commit from previous open commit https://github.com/pytorch/audio/issues/2267. Organizes seed to utils.

Pull Request resolved: https://github.com/pytorch/audio/pull/2425

Reviewed By: carolineechen, nateanl

Differential Revision: D36787599

Pulled By: skim0514

fbshipit-source-id: 37a0d632d13d4336a830c4b98bdb04828ed88c20

test/torchaudio_unittest/common_utils/case_utils.py

@@ -97,6 +97,7 @@ class TestBaseMixin:
     def setUp(self):
         super().setUp()
         set_audio_backend(self.backend)
+        torch.random.manual_seed(2434)
 
     @property
     def complex_dtype(self):

test/torchaudio_unittest/functional/autograd_impl.py

@@ -33,7 +33,6 @@ class Autograd(TestBaseMixin):
         assert gradgradcheck(transform, inputs_)
 
     def test_lfilter_x(self):
-        torch.random.manual_seed(2434)
         x = get_whitenoise(sample_rate=22050, duration=0.01, n_channels=2)
         a = torch.tensor([0.7, 0.2, 0.6])
         b = torch.tensor([0.4, 0.2, 0.9])
@@ -41,7 +40,6 @@ class Autograd(TestBaseMixin):
         self.assert_grad(F.lfilter, (x, a, b), enable_all_grad=False)
 
     def test_lfilter_a(self):
-        torch.random.manual_seed(2434)
         x = get_whitenoise(sample_rate=22050, duration=0.01, n_channels=2)
         a = torch.tensor([0.7, 0.2, 0.6])
         b = torch.tensor([0.4, 0.2, 0.9])
@@ -49,7 +47,6 @@ class Autograd(TestBaseMixin):
         self.assert_grad(F.lfilter, (x, a, b), enable_all_grad=False)
 
     def test_lfilter_b(self):
-        torch.random.manual_seed(2434)
         x = get_whitenoise(sample_rate=22050, duration=0.01, n_channels=2)
         a = torch.tensor([0.7, 0.2, 0.6])
         b = torch.tensor([0.4, 0.2, 0.9])
@@ -57,28 +54,24 @@ class Autograd(TestBaseMixin):
         self.assert_grad(F.lfilter, (x, a, b), enable_all_grad=False)
 
     def test_lfilter_all_inputs(self):
-        torch.random.manual_seed(2434)
         x = get_whitenoise(sample_rate=22050, duration=0.01, n_channels=2)
         a = torch.tensor([0.7, 0.2, 0.6])
         b = torch.tensor([0.4, 0.2, 0.9])
         self.assert_grad(F.lfilter, (x, a, b))
 
     def test_lfilter_filterbanks(self):
-        torch.random.manual_seed(2434)
         x = get_whitenoise(sample_rate=22050, duration=0.01, n_channels=3)
         a = torch.tensor([[0.7, 0.2, 0.6], [0.8, 0.2, 0.9]])
         b = torch.tensor([[0.4, 0.2, 0.9], [0.7, 0.2, 0.6]])
         self.assert_grad(partial(F.lfilter, batching=False), (x, a, b))
 
     def test_lfilter_batching(self):
-        torch.random.manual_seed(2434)
         x = get_whitenoise(sample_rate=22050, duration=0.01, n_channels=2)
         a = torch.tensor([[0.7, 0.2, 0.6], [0.8, 0.2, 0.9]])
         b = torch.tensor([[0.4, 0.2, 0.9], [0.7, 0.2, 0.6]])
         self.assert_grad(F.lfilter, (x, a, b))
 
     def test_filtfilt_a(self):
-        torch.random.manual_seed(2434)
         x = get_whitenoise(sample_rate=22050, duration=0.01, n_channels=2)
         a = torch.tensor([0.7, 0.2, 0.6])
         b = torch.tensor([0.4, 0.2, 0.9])
@@ -86,7 +79,6 @@ class Autograd(TestBaseMixin):
         self.assert_grad(F.filtfilt, (x, a, b), enable_all_grad=False)
 
     def test_filtfilt_b(self):
-        torch.random.manual_seed(2434)
         x = get_whitenoise(sample_rate=22050, duration=0.01, n_channels=2)
         a = torch.tensor([0.7, 0.2, 0.6])
         b = torch.tensor([0.4, 0.2, 0.9])
@@ -94,21 +86,18 @@ class Autograd(TestBaseMixin):
         self.assert_grad(F.filtfilt, (x, a, b), enable_all_grad=False)
 
     def test_filtfilt_all_inputs(self):
-        torch.random.manual_seed(2434)
         x = get_whitenoise(sample_rate=22050, duration=0.01, n_channels=2)
         a = torch.tensor([0.7, 0.2, 0.6])
         b = torch.tensor([0.4, 0.2, 0.9])
         self.assert_grad(F.filtfilt, (x, a, b))
 
     def test_filtfilt_batching(self):
-        torch.random.manual_seed(2434)
         x = get_whitenoise(sample_rate=22050, duration=0.01, n_channels=2)
         a = torch.tensor([[0.7, 0.2, 0.6], [0.8, 0.2, 0.9]])
         b = torch.tensor([[0.4, 0.2, 0.9], [0.7, 0.2, 0.6]])
         self.assert_grad(F.filtfilt, (x, a, b))
 
     def test_biquad(self):
-        torch.random.manual_seed(2434)
         x = get_whitenoise(sample_rate=22050, duration=0.01, n_channels=1)
         a = torch.tensor([0.7, 0.2, 0.6])
         b = torch.tensor([0.4, 0.2, 0.9])
@@ -121,7 +110,6 @@ class Autograd(TestBaseMixin):
         ]
     )
     def test_band_biquad(self, central_freq, Q, noise):
-        torch.random.manual_seed(2434)
         sr = 22050
         x = get_whitenoise(sample_rate=sr, duration=0.01, n_channels=1)
         central_freq = torch.tensor(central_freq)
@@ -135,7 +123,6 @@ class Autograd(TestBaseMixin):
         ]
     )
     def test_bass_biquad(self, central_freq, Q, gain):
-        torch.random.manual_seed(2434)
         sr = 22050
         x = get_whitenoise(sample_rate=sr, duration=0.01, n_channels=1)
         central_freq = torch.tensor(central_freq)
@@ -150,7 +137,6 @@ class Autograd(TestBaseMixin):
         ]
     )
     def test_treble_biquad(self, central_freq, Q, gain):
-        torch.random.manual_seed(2434)
         sr = 22050
         x = get_whitenoise(sample_rate=sr, duration=0.01, n_channels=1)
         central_freq = torch.tensor(central_freq)
@@ -167,7 +153,6 @@ class Autograd(TestBaseMixin):
         ]
     )
     def test_allpass_biquad(self, central_freq, Q):
-        torch.random.manual_seed(2434)
         sr = 22050
         x = get_whitenoise(sample_rate=sr, duration=0.01, n_channels=1)
         central_freq = torch.tensor(central_freq)
@@ -183,7 +168,6 @@ class Autograd(TestBaseMixin):
         ]
     )
     def test_lowpass_biquad(self, cutoff_freq, Q):
-        torch.random.manual_seed(2434)
         sr = 22050
         x = get_whitenoise(sample_rate=sr, duration=0.01, n_channels=1)
         cutoff_freq = torch.tensor(cutoff_freq)
@@ -199,7 +183,6 @@ class Autograd(TestBaseMixin):
         ]
     )
     def test_highpass_biquad(self, cutoff_freq, Q):
-        torch.random.manual_seed(2434)
         sr = 22050
         x = get_whitenoise(sample_rate=sr, duration=0.01, n_channels=1)
         cutoff_freq = torch.tensor(cutoff_freq)
@@ -213,7 +196,6 @@ class Autograd(TestBaseMixin):
         ]
     )
     def test_bandpass_biquad(self, central_freq, Q, const_skirt_gain):
-        torch.random.manual_seed(2434)
         sr = 22050
         x = get_whitenoise(sample_rate=sr, duration=0.01, n_channels=1)
         central_freq = torch.tensor(central_freq)
@@ -227,7 +209,6 @@ class Autograd(TestBaseMixin):
         ]
     )
     def test_equalizer_biquad(self, central_freq, Q, gain):
-        torch.random.manual_seed(2434)
         sr = 22050
         x = get_whitenoise(sample_rate=sr, duration=0.01, n_channels=1)
         central_freq = torch.tensor(central_freq)
@@ -244,7 +225,6 @@ class Autograd(TestBaseMixin):
         ]
     )
     def test_bandreject_biquad(self, central_freq, Q):
-        torch.random.manual_seed(2434)
         sr = 22050
         x = get_whitenoise(sample_rate=sr, duration=0.01, n_channels=1)
         central_freq = torch.tensor(central_freq)
@@ -285,7 +265,6 @@ class Autograd(TestBaseMixin):
         ]
     )
     def test_psd(self, use_mask):
-        torch.random.manual_seed(2434)
         specgram = torch.rand(4, 10, 5, dtype=torch.cfloat)
         if use_mask:
             mask = torch.rand(10, 5)
@@ -294,7 +273,6 @@ class Autograd(TestBaseMixin):
         self.assert_grad(F.psd, (specgram, mask))
 
     def test_mvdr_weights_souden(self):
-        torch.random.manual_seed(2434)
         channel = 4
         n_fft_bin = 5
         psd_speech = torch.rand(n_fft_bin, channel, channel, dtype=torch.cfloat)
@@ -302,7 +280,6 @@ class Autograd(TestBaseMixin):
         self.assert_grad(F.mvdr_weights_souden, (psd_speech, psd_noise, 0))
 
     def test_mvdr_weights_souden_with_tensor(self):
-        torch.random.manual_seed(2434)
         channel = 4
         n_fft_bin = 5
         psd_speech = torch.rand(n_fft_bin, channel, channel, dtype=torch.cfloat)
@@ -312,7 +289,6 @@ class Autograd(TestBaseMixin):
         self.assert_grad(F.mvdr_weights_souden, (psd_speech, psd_noise, reference_channel))
 
     def test_mvdr_weights_rtf(self):
-        torch.random.manual_seed(2434)
         batch_size = 2
         channel = 4
         n_fft_bin = 10
@@ -321,7 +297,6 @@ class Autograd(TestBaseMixin):
         self.assert_grad(F.mvdr_weights_rtf, (rtf, psd_noise, 0))
 
     def test_mvdr_weights_rtf_with_tensor(self):
-        torch.random.manual_seed(2434)
         batch_size = 2
         channel = 4
         n_fft_bin = 10
@@ -338,7 +313,6 @@ class Autograd(TestBaseMixin):
         ]
     )
     def test_rtf_power(self, n_iter, diagonal_loading):
-        torch.random.manual_seed(2434)
         channel = 4
         n_fft_bin = 5
         psd_speech = torch.rand(n_fft_bin, channel, channel, dtype=torch.cfloat)
@@ -352,7 +326,6 @@ class Autograd(TestBaseMixin):
         ]
     )
     def test_rtf_power_with_tensor(self, n_iter, diagonal_loading):
-        torch.random.manual_seed(2434)
         channel = 4
         n_fft_bin = 5
         psd_speech = torch.rand(n_fft_bin, channel, channel, dtype=torch.cfloat)
@@ -362,7 +335,6 @@ class Autograd(TestBaseMixin):
         self.assert_grad(F.rtf_power, (psd_speech, psd_noise, reference_channel, n_iter, diagonal_loading))
 
     def test_apply_beamforming(self):
-        torch.random.manual_seed(2434)
         sr = 8000
         n_fft = 400
         batch_size, num_channels = 2, 3

test/torchaudio_unittest/functional/functional_impl.py

@@ -62,7 +62,6 @@ class Functional(TestBaseMixin):
         The output should be same as the input but shifted
         """
 
-        torch.random.manual_seed(42)
         waveform = torch.rand(2, 44100 * 1, dtype=self.dtype, device=self.device)
         b_coeffs = torch.tensor([0, 0, 0, 1], dtype=self.dtype, device=self.device)
         a_coeffs = torch.tensor([1, 0, 0, 0], dtype=self.dtype, device=self.device)
@@ -115,7 +114,6 @@ class Functional(TestBaseMixin):
         ]
     )
     def test_lfilter_shape(self, input_shape, coeff_shape, target_shape):
-        torch.random.manual_seed(42)
         waveform = torch.rand(*input_shape, dtype=self.dtype, device=self.device)
         b_coeffs = torch.rand(*coeff_shape, dtype=self.dtype, device=self.device)
         a_coeffs = torch.rand(*coeff_shape, dtype=self.dtype, device=self.device)
@@ -283,7 +281,6 @@ class Functional(TestBaseMixin):
         ref = 1.0
         db_mult = math.log10(max(amin, ref))
 
-        torch.manual_seed(0)
         spec = torch.rand(*shape, dtype=self.dtype, device=self.device) * 200
 
         # Spectrogram amplitude -> DB -> amplitude
@@ -307,7 +304,6 @@ class Functional(TestBaseMixin):
         db_mult = math.log10(max(amin, ref))
         top_db = 40.0
 
-        torch.manual_seed(0)
         # A random tensor is used for increased entropy, but the max and min for
         # each spectrogram still need to be predictable. The max determines the
         # decibel cutoff, and the distance from the min must be large enough
@@ -333,7 +329,6 @@ class Functional(TestBaseMixin):
         list(itertools.product([(2, 1025, 400), (1, 201, 100)], [100], [0.0, 30.0], [1, 2], [0.33, 1.0]))
     )
     def test_mask_along_axis(self, shape, mask_param, mask_value, axis, p):
-        torch.random.manual_seed(42)
         specgram = torch.randn(*shape, dtype=self.dtype, device=self.device)
 
         if p != 1.0:
@@ -355,7 +350,6 @@ class Functional(TestBaseMixin):
 
     @parameterized.expand(list(itertools.product([100], [0.0, 30.0], [2, 3], [0.2, 1.0])))
     def test_mask_along_axis_iid(self, mask_param, mask_value, axis, p):
-        torch.random.manual_seed(42)
         specgrams = torch.randn(4, 2, 1025, 400, dtype=self.dtype, device=self.device)
 
         if p != 1.0:
@@ -381,7 +375,6 @@ class Functional(TestBaseMixin):
         Test is run 5 times to bound the probability of no masking occurring to 1e-10
         See https://github.com/pytorch/audio/issues/1478
         """
-        torch.random.manual_seed(42)
         for _ in range(5):
             specgram = torch.randn(*shape, dtype=self.dtype, device=self.device)
             specgram_copy = specgram.clone()
@@ -396,7 +389,6 @@ class Functional(TestBaseMixin):
         Test is run 5 times to bound the probability of no masking occurring to 1e-10
         See https://github.com/pytorch/audio/issues/1478
         """
-        torch.random.manual_seed(42)
         for _ in range(5):
             specgrams = torch.randn(4, 2, 1025, 400, dtype=self.dtype, device=self.device)
             specgrams_copy = specgrams.clone()
@@ -478,7 +470,6 @@ class Functional(TestBaseMixin):
         num_frames = 400
         batch_size = 2
 
-        torch.random.manual_seed(42)
         spec = torch.randn(batch_size, num_freq, num_frames, dtype=self.complex_dtype, device=self.device)
 
         phase_advance = torch.linspace(0, np.pi * hop_length, num_freq, dtype=self.dtype, device=self.device)[..., None]
@@ -534,7 +525,6 @@ class Functional(TestBaseMixin):
     )
     def test_pitch_shift_shape(self, n_steps):
         sample_rate = 16000
-        torch.random.manual_seed(42)
         waveform = torch.rand(2, 44100 * 1, dtype=self.dtype, device=self.device)
         waveform_shift = F.pitch_shift(waveform, sample_rate, n_steps)
         assert waveform.size() == waveform_shift.size()

test/torchaudio_unittest/models/wav2vec2/fairseq_integration_test.py

@@ -207,6 +207,7 @@ class TestFairseqIntegration(TorchaudioTestCase):
         imported = import_fairseq_model(original).eval()
 
         # Without mask
+        torch.manual_seed(0)
         x = torch.randn(batch_size, num_frames)
         ref = original(x, torch.zeros_like(x))["encoder_out"].transpose(0, 1)
         hyp, _ = imported(x)

test/torchaudio_unittest/transforms/autograd_test_impl.py

@@ -259,7 +259,7 @@ class AutogradTestMixin(TestBaseMixin):
         spectrogram = get_spectrogram(waveform, n_fft=n_fft, power=None)
 
         # 1e-3 is too small (on CPU)
-        epsilon = 1e-2
+        epsilon = 2e-2
         too_close = spectrogram.abs() < epsilon
         spectrogram[too_close] = epsilon * spectrogram[too_close] / spectrogram[too_close].abs()
         self.assert_grad(transform, [spectrogram])