Compute deltas

test/test_compliance_kaldi.py

@@ -319,5 +319,6 @@ def test_resample_waveform_multi_channel(self):
             single_channel_sampled = kaldi.resample_waveform(single_channel, sample_rate, sample_rate // 2)
             self.assertTrue(torch.allclose(multi_sound_sampled[i, :], single_channel_sampled, rtol=1e-4))
 
+
 if __name__ == '__main__':
     unittest.main()

test/test_functional.py

@@ -18,6 +18,32 @@
 class TestFunctional(unittest.TestCase):
     data_sizes = [(2, 20), (3, 15), (4, 10)]
     number_of_trials = 100
+    specgram = torch.tensor([1., 2., 3., 4.])
+
+    def _test_compute_deltas(self, specgram, expected, win_length=3, atol=1e-6, rtol=1e-8):
+        computed = F.compute_deltas(specgram, win_length=win_length)
+        self.assertTrue(computed.shape == expected.shape, (computed.shape, expected.shape))
+        torch.testing.assert_allclose(computed, expected, atol=atol, rtol=rtol)
+
+    def test_compute_deltas_onechannel(self):
+        specgram = self.specgram.unsqueeze(0).unsqueeze(0)
+        expected = torch.tensor([[[0.5, 1.0, 1.0, 0.5]]])
+        self._test_compute_deltas(specgram, expected)
+
+    def test_compute_deltas_twochannel(self):
+        specgram = self.specgram.repeat(1, 2, 1)
+        expected = torch.tensor([[[0.5, 1.0, 1.0, 0.5],
+                                  [0.5, 1.0, 1.0, 0.5]]])
+        self._test_compute_deltas(specgram, expected)
+
+    def test_compute_deltas_randn(self):
+        channel = 13
+        n_mfcc = channel * 3
+        time = 1021
+        win_length = 2 * 7 + 1
+        specgram = torch.randn(channel, n_mfcc, time)
+        computed = F.compute_deltas(specgram, win_length=win_length)
+        self.assertTrue(computed.shape == specgram.shape, (computed.shape, specgram.shape))
 
     def _compare_estimate(self, sound, estimate, atol=1e-6, rtol=1e-8):
         # trim sound for case when constructed signal is shorter than original

test/test_transforms.py

@@ -4,8 +4,9 @@
 
 import torch
 import torchaudio
-from torchaudio.common_utils import IMPORT_LIBROSA, IMPORT_SCIPY
 import torchaudio.transforms as transforms
+import torchaudio.functional as F
+from torchaudio.common_utils import IMPORT_LIBROSA, IMPORT_SCIPY
 import unittest
 import common_utils
 
@@ -281,5 +282,37 @@ def test_resample_size(self):
         # we expect the downsampled signal to have half as many samples
         self.assertTrue(down_sampled.size(-1) == waveform.size(-1) // 2)
 
+    def test_compute_deltas(self):
+        channel = 13
+        n_mfcc = channel * 3
+        time = 1021
+        win_length = 2 * 7 + 1
+        specgram = torch.randn(channel, n_mfcc, time)
+        transform = transforms.ComputeDeltas(win_length=win_length)
+        computed = transform(specgram)
+        self.assertTrue(computed.shape == specgram.shape, (computed.shape, specgram.shape))
+
+    def test_compute_deltas_transform_same_as_functional(self, atol=1e-6, rtol=1e-8):
+        channel = 13
+        n_mfcc = channel * 3
+        time = 1021
+        win_length = 2 * 7 + 1
+        specgram = torch.randn(channel, n_mfcc, time)
+
+        transform = transforms.ComputeDeltas(win_length=win_length)
+        computed_transform = transform(specgram)
+
+        computed_functional = F.compute_deltas(specgram, win_length=win_length)
+        torch.testing.assert_allclose(computed_functional, computed_transform, atol=atol, rtol=rtol)
+
+    def test_compute_deltas_twochannel(self):
+        specgram = torch.tensor([1., 2., 3., 4.]).repeat(1, 2, 1)
+        expected = torch.tensor([[[0.5, 1.0, 1.0, 0.5],
+                                  [0.5, 1.0, 1.0, 0.5]]])
+        transform = transforms.ComputeDeltas()
+        computed = transform(specgram)
+        self.assertTrue(computed.shape == specgram.shape, (computed.shape, specgram.shape))
+
+
 if __name__ == '__main__':
     unittest.main()

torchaudio/functional.py

@@ -20,6 +20,7 @@
     "biquad",
 ]
 
+
 # TODO: remove this once https://github.com/pytorch/pytorch/issues/21478 gets solved
 @torch.jit.ignore
 def _stft(
@@ -652,3 +653,50 @@ def lowpass_biquad(waveform, sample_rate, cutoff_freq, Q=0.707):
     a1 = -2 * math.cos(w0)
     a2 = 1 - alpha
     return biquad(waveform, b0, b1, b2, a0, a1, a2)
+
+
+def compute_deltas(specgram, win_length=5, mode="replicate"):
+    # type: (Tensor, int, str) -> Tensor
+    r"""Compute delta coefficients of a tensor, usually a spectrogram:
+
+    .. math::
+        d_t = \frac{\sum_{n=1}^{\text{N}} n (c_{t+n} - c_{t-n})}{2 \sum_{n=1}^{\text{N} n^2}
+
+    where :math:`d_t` is the deltas at time :math:`t`,
+    :math:`c_t` is the spectrogram coeffcients at time :math:`t`,
+    :math:`N` is (`win_length`-1)//2.
+
+    Args:
+        specgram (torch.Tensor): Tensor of audio of dimension (channel, n_mfcc, time)
+        win_length (int): The window length used for computing delta
+        mode (str): Mode parameter passed to padding
+
+    Returns:
+        deltas (torch.Tensor): Tensor of audio of dimension (channel, n_mfcc, time)
+
+    Example
+        >>> specgram = torch.randn(1, 40, 1000)
+        >>> delta = compute_deltas(specgram)
+        >>> delta2 = compute_deltas(delta)
+    """
+
+    assert win_length >= 3
+    assert specgram.dim() == 3
+    assert not specgram.shape[1] % specgram.shape[0]
+
+    n = (win_length - 1) // 2
+
+    # twice sum of integer squared
+    denom = n * (n + 1) * (2 * n + 1) / 3
+
+    specgram = torch.nn.functional.pad(specgram, (n, n), mode=mode)
+
+    kernel = (
+        torch
+        .arange(-n, n + 1, 1, device=specgram.device, dtype=specgram.dtype)
+        .repeat(specgram.shape[1], specgram.shape[0], 1)
+    )
+
+    return torch.nn.functional.conv1d(
+        specgram, kernel, groups=specgram.shape[1] // specgram.shape[0]
+    ) / denom

torchaudio/transforms.py

@@ -365,3 +365,30 @@ def forward(self, waveform):
             return kaldi.resample_waveform(waveform, self.orig_freq, self.new_freq)
 
         raise ValueError('Invalid resampling method: %s' % (self.resampling_method))
+
+
+class ComputeDeltas(torch.jit.ScriptModule):
+    r"""Compute delta coefficients of a tensor, usually a spectrogram.
+
+    See `torchaudio.functional.compute_deltas` for more details.
+
+    Args:
+        win_length (int): The window length used for computing delta.
+    """
+    __constants__ = ['win_length']
+
+    def __init__(self, win_length=5, mode="replicate"):
+        super(ComputeDeltas, self).__init__()
+        self.win_length = win_length
+        self.mode = torch.jit.Attribute(mode, str)
+
+    @torch.jit.script_method
+    def forward(self, specgram):
+        r"""
+        Args:
+            specgram (torch.Tensor): Tensor of audio of dimension (channel, n_mfcc, time)
+
+        Returns:
+            deltas (torch.Tensor): Tensor of audio of dimension (channel, n_mfcc, time)
+        """
+        return F.compute_deltas(specgram, win_length=self.win_length, mode=self.mode)