reimplement in torchscriptable way

test/torchaudio_unittest/functional/torchscript_consistency_impl.py

@@ -592,12 +592,17 @@ def func(tensor):
         self._assert_consistency(func, tensor)
 
     def test_resample(self):
-        def func(tensor):
+        def func_sinc(tensor):
+            sr1, sr2 = 16000., 8000.
+            return F.resample(tensor, sr1, sr2, resampling_method="sinc_interpolation")
+
+        def func_kaiser(tensor):
             sr1, sr2 = 16000., 8000.
-            return F.resample(tensor, sr1, sr2)
+            return F.resample(tensor, sr1, sr2, resampling_method="kaiser_window")
 
         tensor = common_utils.get_whitenoise(sample_rate=16000)
-        self._assert_consistency(func, tensor)
+        self._assert_consistency(func_sinc, tensor)
+        self._assert_consistency(func_kaiser, tensor)
 
     @parameterized.expand([(True, ), (False, )])
     def test_phase_vocoder(self, test_paseudo_complex):

torchaudio/functional/functional.py

@@ -1305,7 +1305,7 @@ def _get_sinc_resample_kernel(
         lowpass_filter_width: int,
         rolloff: float,
         resampling_method: str,
-        beta: Optional[float],
+        beta: float,
         device: torch.device = torch.device("cpu"),
         dtype: Optional[torch.dtype] = None):
 
@@ -1328,9 +1328,6 @@ def _get_sinc_resample_kernel(
     orig_freq = int(orig_freq) // gcd
     new_freq = int(new_freq) // gcd
 
-    if resampling_method == "kaiser_window" and beta is None:
-        beta = 14.769656459379492
-
     assert lowpass_filter_width > 0
     kernels = []
     base_freq = min(orig_freq, new_freq)
@@ -1373,9 +1370,10 @@ def _get_sinc_resample_kernel(
         # at specific positions, not over a regular grid.
         if resampling_method == "sinc_interpolation":
             window = torch.cos(t * math.pi / lowpass_filter_width / 2)**2
-        elif resampling_method == "kaiser_window":
-            beta = torch.tensor(beta, dtype=float)
-            window = torch.i0(beta * torch.sqrt(1 - (t / lowpass_filter_width) ** 2)) / torch.i0(beta)
+        else:
+            # kaiser_window
+            beta_tensor = torch.tensor(beta)
+            window = torch.i0(beta_tensor * torch.sqrt(1 - (t / lowpass_filter_width) ** 2)) / torch.i0(beta_tensor)
         t *= math.pi
         kernel = torch.where(t == 0, torch.tensor(1.).to(t), torch.sin(t) / t)
         kernel.mul_(window)
@@ -1422,7 +1420,7 @@ def resample(
         lowpass_filter_width: int = 6,
         rolloff: float = 0.99,
         resampling_method: str = "sinc_interpolation",
-        beta: Optional[float] = None,
+        beta: float = 14.769656459379492,
 ) -> Tensor:
     r"""Resamples the waveform at the new frequency. This matches Kaldi's OfflineFeatureTpl ResampleWaveform
     which uses a LinearResample (resample a signal at linearly spaced intervals to upsample/downsample

torchaudio/transforms.py

@@ -686,7 +686,7 @@ def __init__(self,
                  resampling_method: str = 'sinc_interpolation',
                  lowpass_filter_width: int = 6,
                  rolloff: float = 0.99,
-                 beta: Optional[float] = None) -> None:
+                 beta: float = 14.769656459379492) -> None:
         super(Resample, self).__init__()
 
         self.orig_freq = orig_freq