Add return_complex to F.spectrogram and T.Spectrogram

test/torchaudio_unittest/transforms/autograd_test_impl.py

@@ -29,6 +29,10 @@ def assert_grad(
         assert gradgradcheck(transform, inputs_, nondet_tol=nondet_tol)
 
     @parameterized.expand([
+        ({'pad': 0, 'normalized': False, 'power': None, 'return_complex': True}, ),
+        ({'pad': 3, 'normalized': False, 'power': None, 'return_complex': True}, ),
+        ({'pad': 0, 'normalized': True, 'power': None, 'return_complex': True}, ),
+        ({'pad': 3, 'normalized': True, 'power': None, 'return_complex': True}, ),
         ({'pad': 0, 'normalized': False, 'power': None}, ),
         ({'pad': 3, 'normalized': False, 'power': None}, ),
         ({'pad': 0, 'normalized': True, 'power': None}, ),

test/torchaudio_unittest/transforms/librosa_compatibility_test.py

@@ -45,6 +45,20 @@ def test_spectrogram(self, n_fft, hop_length, power):
         out_torch = spect_transform(sound).squeeze().cpu()
         self.assertEqual(out_torch, torch.from_numpy(out_librosa), atol=1e-5, rtol=1e-5)
 
+    def test_spectrogram_complex(self):
+        n_fft = 400
+        hop_length = 200
+        sample_rate = 16000
+        sound = common_utils.get_sinusoid(n_channels=1, sample_rate=sample_rate)
+        sound_librosa = sound.cpu().numpy().squeeze()
+        spect_transform = torchaudio.transforms.Spectrogram(
+            n_fft=n_fft, hop_length=hop_length, power=None, return_complex=True)
+        out_librosa, _ = librosa.core.spectrum._spectrogram(
+            y=sound_librosa, n_fft=n_fft, hop_length=hop_length, power=1)
+
+        out_torch = spect_transform(sound).squeeze()
+        self.assertEqual(out_torch.abs(), torch.from_numpy(out_librosa), atol=1e-5, rtol=1e-5)
+
     @parameterized.expand([
         param(norm=norm, mel_scale=mel_scale, **p.kwargs)
         for p in [

test/torchaudio_unittest/transforms/torchscript_consistency_impl.py

@@ -25,6 +25,10 @@ def test_Spectrogram(self):
         tensor = torch.rand((1, 1000))
         self._assert_consistency(T.Spectrogram(), tensor)
 
+    def test_Spectrogram_return_complex(self):
+        tensor = torch.rand((1, 1000))
+        self._assert_consistency(T.Spectrogram(power=None, return_complex=True), tensor)
+
     @skipIfRocm
     def test_GriffinLim(self):
         tensor = torch.rand((1, 201, 6))

torchaudio/functional/functional.py

@@ -48,7 +48,8 @@ def spectrogram(
         normalized: bool,
         center: bool = True,
         pad_mode: str = "reflect",
-        onesided: bool = True
+        onesided: bool = True,
+        return_complex: bool = False,
 ) -> Tensor:
     r"""Create a spectrogram or a batch of spectrograms from a raw audio signal.
     The spectrogram can be either magnitude-only or complex.
@@ -71,12 +72,22 @@ def spectrogram(
             :attr:`center` is ``True``. Default: ``"reflect"``
         onesided (bool, optional): controls whether to return half of results to
             avoid redundancy. Default: ``True``
+        return_complex (bool, optional):
+            ``return_complex = True``, this function returns the resulting Tensor in
+            complex dtype, otherwise it returns the resulting Tensor in real dtype with extra
+            dimension for real and imaginary parts. (see ``torch.view_as_real``).
+            When ``power`` is provided, the value must be False, as the resulting
+            Tensor represents real-valued power.
 
     Returns:
         Tensor: Dimension (..., freq, time), freq is
         ``n_fft // 2 + 1`` and ``n_fft`` is the number of
         Fourier bins, and time is the number of window hops (n_frame).
     """
+    if power is not None and return_complex:
+        raise ValueError(
+            'When `power` is provided, the return value is real-valued. '
+            'Therefore, `return_complex` must be False.')
 
     if pad > 0:
         # TODO add "with torch.no_grad():" back when JIT supports it
@@ -109,7 +120,9 @@ def spectrogram(
         if power == 1.0:
             return spec_f.abs()
         return spec_f.abs().pow(power)
-    return torch.view_as_real(spec_f)
+    if not return_complex:
+        return torch.view_as_real(spec_f)
+    return spec_f
 
 
 def griffinlim(

torchaudio/transforms.py

@@ -52,6 +52,12 @@ class Spectrogram(torch.nn.Module):
             :attr:`center` is ``True``. Default: ``"reflect"``
         onesided (bool, optional): controls whether to return half of results to
             avoid redundancy Default: ``True``
+        return_complex (bool, optional):
+            ``return_complex = True``, this function returns the resulting Tensor in
+            complex dtype, otherwise it returns the resulting Tensor in real dtype with extra
+            dimension for real and imaginary parts. (see ``torch.view_as_real``).
+            When ``power`` is provided, the value must be False, as the resulting
+            Tensor represents real-valued power.
     """
     __constants__ = ['n_fft', 'win_length', 'hop_length', 'pad', 'power', 'normalized']
 
@@ -66,7 +72,8 @@ def __init__(self,
                  wkwargs: Optional[dict] = None,
                  center: bool = True,
                  pad_mode: str = "reflect",
-                 onesided: bool = True) -> None:
+                 onesided: bool = True,
+                 return_complex: bool = False) -> None:
         super(Spectrogram, self).__init__()
         self.n_fft = n_fft
         # number of FFT bins. the returned STFT result will have n_fft // 2 + 1
@@ -81,6 +88,7 @@ def __init__(self,
         self.center = center
         self.pad_mode = pad_mode
         self.onesided = onesided
+        self.return_complex = return_complex
 
     def forward(self, waveform: Tensor) -> Tensor:
         r"""
@@ -103,7 +111,8 @@ def forward(self, waveform: Tensor) -> Tensor:
             self.normalized,
             self.center,
             self.pad_mode,
-            self.onesided
+            self.onesided,
+            self.return_complex,
         )