Modifying modules and functions to support JIT and cuda (#118)

build_tools/travis/install.sh

@@ -52,8 +52,8 @@ pip install -r requirements.txt
 
  # Install the following only if running tests
 if [[ "$SKIP_TESTS" != "true" ]]; then
-    # PyTorch
-    conda install --yes pytorch -c pytorch
+    # PyTorch (nightly as 1.1 does not have Optional for type annotations)
+    conda install --yes pytorch-nightly-cpu -c pytorch
 
      # TorchAudio CPP Extensions
     pip install .

test/test_jit.py

+import torch
+import torchaudio.functional as F
+import torchaudio.transforms as transforms
+import unittest
+
+RUN_CUDA = torch.cuda.is_available()
+print('Run test with cuda:', RUN_CUDA)
+
+
+class Test_JIT(unittest.TestCase):
+    def _get_script_module(self, f, *args):
+        # takes a transform function `f` and wraps it in a script module
+        class MyModule(torch.jit.ScriptModule):
+            def __init__(self):
+                super(MyModule, self).__init__()
+                self.module = f(*args)
+                self.module.eval()
+
+            @torch.jit.script_method
+            def forward(self, tensor):
+                return self.module(tensor)
+
+        return MyModule()
+
+    def _test_script_module(self, tensor, f, *args):
+        # tests a script module that wraps a transform function `f` by feeding
+        # the tensor into the forward function
+        jit_out = self._get_script_module(f, *args).cuda()(tensor)
+        py_out = f(*args).cuda()(tensor)
+
+        self.assertTrue(torch.allclose(jit_out, py_out))
+
+    def test_torchscript_scale(self):
+        @torch.jit.script
+        def jit_method(tensor, factor):
+            # type: (Tensor, int) -> Tensor
+            return F.scale(tensor, factor)
+
+        tensor = torch.rand((10, 1))
+        factor = 2
+
+        jit_out = jit_method(tensor, factor)
+        py_out = F.scale(tensor, factor)
+
+        self.assertTrue(torch.allclose(jit_out, py_out))
+
+    @unittest.skipIf(not RUN_CUDA, "no CUDA")
+    def test_scriptmodule_scale(self):
+        tensor = torch.rand((10, 1), device="cuda")
+
+        self._test_script_module(tensor, transforms.Scale)
+
+    def test_torchscript_pad_trim(self):
+        @torch.jit.script
+        def jit_method(tensor, ch_dim, max_len, len_dim, fill_value):
+            # type: (Tensor, int, int, int, float) -> Tensor
+            return F.pad_trim(tensor, ch_dim, max_len, len_dim, fill_value)
+
+        tensor = torch.rand((10, 1))
+        ch_dim = 1
+        max_len = 5
+        len_dim = 0
+        fill_value = 3.
+
+        jit_out = jit_method(tensor, ch_dim, max_len, len_dim, fill_value)
+        py_out = F.pad_trim(tensor, ch_dim, max_len, len_dim, fill_value)
+
+        self.assertTrue(torch.allclose(jit_out, py_out))
+
+    @unittest.skipIf(not RUN_CUDA, "no CUDA")
+    def test_scriptmodule_pad_trim(self):
+        tensor = torch.rand((1, 10), device="cuda")
+        max_len = 5
+
+        self._test_script_module(tensor, transforms.PadTrim, max_len)
+
+    def test_torchscript_downmix_mono(self):
+        @torch.jit.script
+        def jit_method(tensor, ch_dim):
+            # type: (Tensor, int) -> Tensor
+            return F.downmix_mono(tensor, ch_dim)
+
+        tensor = torch.rand((10, 1))
+        ch_dim = 1
+
+        jit_out = jit_method(tensor, ch_dim)
+        py_out = F.downmix_mono(tensor, ch_dim)
+
+        self.assertTrue(torch.allclose(jit_out, py_out))
+
+    @unittest.skipIf(not RUN_CUDA, "no CUDA")
+    def test_scriptmodule_downmix_mono(self):
+        tensor = torch.rand((1, 10), device="cuda")
+
+        self._test_script_module(tensor, transforms.DownmixMono)
+
+    def test_torchscript_LC2CL(self):
+        @torch.jit.script
+        def jit_method(tensor):
+            # type: (Tensor) -> Tensor
+            return F.LC2CL(tensor)
+
+        tensor = torch.rand((10, 1))
+
+        jit_out = jit_method(tensor)
+        py_out = F.LC2CL(tensor)
+
+        self.assertTrue(torch.allclose(jit_out, py_out))
+
+    @unittest.skipIf(not RUN_CUDA, "no CUDA")
+    def test_scriptmodule_LC2CL(self):
+        tensor = torch.rand((10, 1), device="cuda")
+
+        self._test_script_module(tensor, transforms.LC2CL)
+
+    def test_torchscript_spectrogram(self):
+        @torch.jit.script
+        def jit_method(sig, pad, window, n_fft, hop, ws, power, normalize):
+            # type: (Tensor, int, Tensor, int, int, int, int, bool) -> Tensor
+            return F.spectrogram(sig, pad, window, n_fft, hop, ws, power, normalize)
+
+        tensor = torch.rand((1, 1000))
+        n_fft = 400
+        ws = 400
+        hop = 200
+        pad = 0
+        window = torch.hann_window(ws)
+        power = 2
+        normalize = False
+
+        jit_out = jit_method(tensor, pad, window, n_fft, hop, ws, power, normalize)
+        py_out = F.spectrogram(tensor, pad, window, n_fft, hop, ws, power, normalize)
+
+        self.assertTrue(torch.allclose(jit_out, py_out))
+
+    @unittest.skipIf(not RUN_CUDA, "no CUDA")
+    def test_scriptmodule_Spectrogram(self):
+        tensor = torch.rand((1, 1000), device="cuda")
+
+        self._test_script_module(tensor, transforms.Spectrogram)
+
+    def test_torchscript_create_fb_matrix(self):
+        @torch.jit.script
+        def jit_method(n_stft, f_min, f_max, n_mels):
+            # type: (int, float, float, int) -> Tensor
+            return F.create_fb_matrix(n_stft, f_min, f_max, n_mels)
+
+        n_stft = 100
+        f_min = 0.
+        f_max = 20.
+        n_mels = 10
+
+        jit_out = jit_method(n_stft, f_min, f_max, n_mels)
+        py_out = F.create_fb_matrix(n_stft, f_min, f_max, n_mels)
+
+        self.assertTrue(torch.allclose(jit_out, py_out))
+
+    @unittest.skipIf(not RUN_CUDA, "no CUDA")
+    def test_scriptmodule_MelScale(self):
+        spec_f = torch.rand((1, 6, 201), device="cuda")
+
+        self._test_script_module(spec_f, transforms.MelScale)
+
+    def test_torchscript_spectrogram_to_DB(self):
+        @torch.jit.script
+        def jit_method(spec, multiplier, amin, db_multiplier, top_db):
+            # type: (Tensor, float, float, float, Optional[float]) -> Tensor
+            return F.spectrogram_to_DB(spec, multiplier, amin, db_multiplier, top_db)
+
+        spec = torch.rand((10, 1))
+        multiplier = 10.
+        amin = 1e-10
+        db_multiplier = 0.
+        top_db = 80.
+
+        jit_out = jit_method(spec, multiplier, amin, db_multiplier, top_db)
+        py_out = F.spectrogram_to_DB(spec, multiplier, amin, db_multiplier, top_db)
+
+        self.assertTrue(torch.allclose(jit_out, py_out))
+
+    @unittest.skipIf(not RUN_CUDA, "no CUDA")
+    def test_scriptmodule_SpectrogramToDB(self):
+        spec = torch.rand((10, 1), device="cuda")
+
+        self._test_script_module(spec, transforms.SpectrogramToDB)
+
+    def test_torchscript_create_dct(self):
+        @torch.jit.script
+        def jit_method(n_mfcc, n_mels, norm):
+            # type: (int, int, Optional[str]) -> Tensor
+            return F.create_dct(n_mfcc, n_mels, norm)
+
+        n_mfcc = 40
+        n_mels = 128
+        norm = 'ortho'
+
+        jit_out = jit_method(n_mfcc, n_mels, norm)
+        py_out = F.create_dct(n_mfcc, n_mels, norm)
+
+        self.assertTrue(torch.allclose(jit_out, py_out))
+
+    @unittest.skipIf(not RUN_CUDA, "no CUDA")
+    def test_scriptmodule_MFCC(self):
+        tensor = torch.rand((1, 1000), device="cuda")
+
+        self._test_script_module(tensor, transforms.MFCC)
+
+    @unittest.skipIf(not RUN_CUDA, "no CUDA")
+    def test_scriptmodule_MelSpectrogram(self):
+        tensor = torch.rand((1, 1000), device="cuda")
+
+        self._test_script_module(tensor, transforms.MelSpectrogram)
+
+    def test_torchscript_BLC2CBL(self):
+        @torch.jit.script
+        def jit_method(tensor):
+            # type: (Tensor) -> Tensor
+            return F.BLC2CBL(tensor)
+
+        tensor = torch.rand((10, 1000, 1))
+
+        jit_out = jit_method(tensor)
+        py_out = F.BLC2CBL(tensor)
+
+        self.assertTrue(torch.allclose(jit_out, py_out))
+
+    @unittest.skipIf(not RUN_CUDA, "no CUDA")
+    def test_scriptmodule_BLC2CBL(self):
+        tensor = torch.rand((10, 1000, 1), device="cuda")
+
+        self._test_script_module(tensor, transforms.BLC2CBL)
+
+    def test_torchscript_mu_law_encoding(self):
+        @torch.jit.script
+        def jit_method(tensor, qc):
+            # type: (Tensor, int) -> Tensor
+            return F.mu_law_encoding(tensor, qc)
+
+        tensor = torch.rand((10, 1))
+        qc = 256
+
+        jit_out = jit_method(tensor, qc)
+        py_out = F.mu_law_encoding(tensor, qc)
+
+        self.assertTrue(torch.allclose(jit_out, py_out))
+
+    @unittest.skipIf(not RUN_CUDA, "no CUDA")
+    def test_scriptmodule_MuLawEncoding(self):
+        tensor = torch.rand((10, 1), device="cuda")
+
+        self._test_script_module(tensor, transforms.MuLawEncoding)
+
+    def test_torchscript_mu_law_expanding(self):
+        @torch.jit.script
+        def jit_method(tensor, qc):
+            # type: (Tensor, int) -> Tensor
+            return F.mu_law_expanding(tensor, qc)
+
+        tensor = torch.rand((10, 1))
+        qc = 256
+
+        jit_out = jit_method(tensor, qc)
+        py_out = F.mu_law_expanding(tensor, qc)
+
+        self.assertTrue(torch.allclose(jit_out, py_out))
+
+    @unittest.skipIf(not RUN_CUDA, "no CUDA")
+    def test_scriptmodule_MuLawExpanding(self):
+        tensor = torch.rand((10, 1), device="cuda")
+
+        self._test_script_module(tensor, transforms.MuLawExpanding)
+
+
+if __name__ == '__main__':
+    unittest.main()

test/test_kaldi_io.py

@@ -5,7 +5,7 @@ import unittest
 import test.common_utils
 
 
-class KaldiIOTest(unittest.TestCase):
+class Test_KaldiIO(unittest.TestCase):
     data1 = [[1, 2, 3], [11, 12, 13], [21, 22, 23]]
     data2 = [[31, 32, 33], [41, 42, 43], [51, 52, 53]]
     test_dirpath, test_dir = test.common_utils.create_temp_assets_dir()

test/test_transforms.py

@@ -36,8 +36,9 @@ class Tester(unittest.TestCase):
         result = transforms.Scale()(audio_orig)
         self.assertTrue(result.min() >= -1. and result.max() <= 1.)
 
-        maxminmax = float(max(abs(audio_orig.min()), abs(audio_orig.max())))
+        maxminmax = max(abs(audio_orig.min()), abs(audio_orig.max())).item()
         result = transforms.Scale(factor=maxminmax)(audio_orig)
+
         self.assertTrue((result.min() == -1. or result.max() == 1.) and
                         result.min() >= -1. and result.max() <= 1.)
 
@@ -99,7 +100,7 @@ class Tester(unittest.TestCase):
         audio_orig = self.sig.clone()
         length_orig = audio_orig.size(0)
         length_new = int(length_orig * 1.2)
-        maxminmax = float(max(abs(audio_orig.min()), abs(audio_orig.max())))
+        maxminmax = max(abs(audio_orig.min()), abs(audio_orig.max())).item()
 
         tset = (transforms.Scale(factor=maxminmax),
                 transforms.PadTrim(max_len=length_new, channels_first=False))

torchaudio/__init__.py

@@ -4,7 +4,7 @@ import os.path
 import torch
 import _torch_sox
 
-from torchaudio import transforms, datasets, sox_effects, legacy
+from torchaudio import transforms, datasets, kaldi_io, sox_effects, legacy
 
 
 def check_input(src):

torchaudio/functional.py

@@ -9,16 +9,15 @@ __all__ = [
     'LC2CL',
     'spectrogram',
     'create_fb_matrix',
-    'mel_scale',
     'spectrogram_to_DB',
     'create_dct',
-    'MFCC',
     'BLC2CBL',
     'mu_law_encoding',
     'mu_law_expanding'
 ]
 
 
+@torch.jit.script
 def scale(tensor, factor):
     # type: (Tensor, int) -> Tensor
     """Scale audio tensor from a 16-bit integer (represented as a FloatTensor)
@@ -32,12 +31,13 @@ def scale(tensor, factor):
     Outputs:
         Tensor: Scaled by the scale factor
     """
-    if not tensor.dtype.is_floating_point:
+    if not tensor.is_floating_point():
         tensor = tensor.to(torch.float32)
 
     return tensor / factor
 
 
+@torch.jit.script
 def pad_trim(tensor, ch_dim, max_len, len_dim, fill_value):
     # type: (Tensor, int, int, int, float) -> Tensor
     """Pad/Trim a 2d-Tensor (Signal or Labels)
@@ -53,20 +53,21 @@ def pad_trim(tensor, ch_dim, max_len, len_dim, fill_value):
         Tensor: Padded/trimmed tensor
     """
     if max_len > tensor.size(len_dim):
-        # tuple of (padding_left, padding_right, padding_top, padding_bottom)
+        # array of [padding_left, padding_right, padding_top, padding_bottom]
         # so pad similar to append (aka only right/bottom) and do not pad
         # the length dimension. assumes equal sizes of padding.
         padding = [max_len - tensor.size(len_dim)
                    if (i % 2 == 1) and (i // 2 != len_dim)
                    else 0
-                   for i in range(4)]
-        with torch.no_grad():
+                   for i in [0, 1, 2, 3]]
+        # TODO add "with torch.no_grad():" back when JIT supports it
         tensor = torch.nn.functional.pad(tensor, padding, "constant", fill_value)
     elif max_len < tensor.size(len_dim):
         tensor = tensor.narrow(len_dim, 0, max_len)
     return tensor
 
 
+@torch.jit.script
 def downmix_mono(tensor, ch_dim):
     # type: (Tensor, int) -> Tensor
     """Downmix any stereo signals to mono.
@@ -78,13 +79,14 @@ def downmix_mono(tensor, ch_dim):
     Outputs:
         Tensor: Mono signal
     """
-    if not tensor.dtype.is_floating_point:
+    if not tensor.is_floating_point():
         tensor = tensor.to(torch.float32)
 
     tensor = torch.mean(tensor, ch_dim, True)
     return tensor
 
 
+@torch.jit.script
 def LC2CL(tensor):
     # type: (Tensor) -> Tensor
     """Permute a 2d tensor from samples (n x c) to (c x n)
@@ -98,6 +100,12 @@ def LC2CL(tensor):
     return tensor.transpose(0, 1).contiguous()
 
 
+def _stft(input, n_fft, hop_length, win_length, window, center, pad_mode, normalized, onesided):
+    # type: (Tensor, int, Optional[int], Optional[int], Optional[Tensor], bool, str, bool, bool) -> Tensor
+    return torch.stft(input, n_fft, hop_length, win_length, window, center, pad_mode, normalized, onesided)
+
+
+@torch.jit.script
 def spectrogram(sig, pad, window, n_fft, hop, ws, power, normalize):
     # type: (Tensor, int, Tensor, int, int, int, int, bool) -> Tensor
     """Create a spectrogram from a raw audio signal
@@ -123,21 +131,20 @@ def spectrogram(sig, pad, window, n_fft, hop, ws, power, normalize):
     assert sig.dim() == 2
 
     if pad > 0:
-        with torch.no_grad():
+        # TODO add "with torch.no_grad():" back when JIT supports it
         sig = torch.nn.functional.pad(sig, (pad, pad), "constant")
-    window = window.to(sig.device)
 
     # default values are consistent with librosa.core.spectrum._spectrogram
-    spec_f = torch.stft(sig, n_fft, hop, ws,
-                        window, center=True,
-                        normalized=False, onesided=True,
-                        pad_mode='reflect').transpose(1, 2)
+    spec_f = _stft(sig, n_fft, hop, ws, window,
+                   True, 'reflect', False, True).transpose(1, 2)
+
     if normalize:
         spec_f /= window.pow(2).sum().sqrt()
     spec_f = spec_f.pow(power).sum(-1)  # get power of "complex" tensor (c, l, n_fft)
     return spec_f
 
 
+@torch.jit.script
 def create_fb_matrix(n_stft, f_min, f_max, n_mels):
     # type: (int, float, float, int) -> Tensor
     """ Create a frequency bin conversion matrix.
@@ -150,57 +157,29 @@ def create_fb_matrix(n_stft, f_min, f_max, n_mels):
 
     Outputs:
         Tensor: triangular filter banks (fb matrix)
-    """
-    def _hertz_to_mel(f):
-        # type: (float) -> Tensor
-        return 2595. * torch.log10(torch.tensor(1.) + (f / 700.))
-
-    def _mel_to_hertz(mel):
-        # type: (Tensor) -> Tensor
-        return 700. * (10**(mel / 2595.) - 1.)
 
+    """
     # get stft freq bins
     stft_freqs = torch.linspace(f_min, f_max, n_stft)
     # calculate mel freq bins
-    m_min = 0. if f_min == 0 else _hertz_to_mel(f_min)
-    m_max = _hertz_to_mel(f_max)
+    # hertz to mel(f) is 2595. * math.log10(1. + (f / 700.))
+    m_min = 0. if f_min == 0 else 2595. * math.log10(1. + (f_min / 700.))
+    m_max = 2595. * math.log10(1. + (f_max / 700.))
     m_pts = torch.linspace(m_min, m_max, n_mels + 2)
-    f_pts = _mel_to_hertz(m_pts)
+    # mel to hertz(mel) is 700. * (10**(mel / 2595.) - 1.)
+    f_pts = 700. * (10**(m_pts / 2595.) - 1.)
     # calculate the difference between each mel point and each stft freq point in hertz
     f_diff = f_pts[1:] - f_pts[:-1]  # (n_mels + 1)
     slopes = f_pts.unsqueeze(0) - stft_freqs.unsqueeze(1)  # (n_stft, n_mels + 2)
     # create overlapping triangles
-    z = torch.tensor(0.)
+    z = torch.zeros(1)
     down_slopes = (-1. * slopes[:, :-2]) / f_diff[:-1]  # (n_stft, n_mels)
     up_slopes = slopes[:, 2:] / f_diff[1:]  # (n_stft, n_mels)
     fb = torch.max(z, torch.min(down_slopes, up_slopes))
     return fb
 
 
-def mel_scale(spec_f, f_min, f_max, n_mels, fb=None):
-    # type: (Tensor, float, float, int, Optional[Tensor]) -> Tuple[Tensor, Tensor]
-    """ This turns a normal STFT into a mel frequency STFT, using a conversion
-    matrix.  This uses triangular filter banks.
-
-    Inputs:
-        spec_f (Tensor): normal STFT
-        f_min (float): minimum frequency
-        f_max (float): maximum frequency
-        n_mels (int): number of mel bins
-        fb (Optional[Tensor]): triangular filter banks (fb matrix)
-
-    Outputs:
-        Tuple[Tensor, Tensor]: triangular filter banks (fb matrix) and mel frequency STFT
-    """
-    if fb is None:
-        fb = create_fb_matrix(spec_f.size(2), f_min, f_max, n_mels).to(spec_f.device)
-    else:
-        # need to ensure same device for dot product
-        fb = fb.to(spec_f.device)
-    spec_m = torch.matmul(spec_f, fb)  # (c, l, n_fft) dot (n_fft, n_mels) -> (c, l, n_mels)
-    return fb, spec_m
-
-
+@torch.jit.script
 def spectrogram_to_DB(spec, multiplier, amin, db_multiplier, top_db=None):
     # type: (Tensor, float, float, float, Optional[float]) -> Tensor
     """Turns a spectrogram from the power/amplitude scale to the decibel scale.
@@ -224,12 +203,15 @@ def spectrogram_to_DB(spec, multiplier, amin, db_multiplier, top_db=None):
     spec_db -= multiplier * db_multiplier
 
     if top_db is not None:
-        spec_db = torch.max(spec_db, spec_db.new_full((1,), spec_db.max() - top_db))
+        new_spec_db_max = torch.tensor(float(spec_db.max()) - top_db, dtype=spec_db.dtype, device=spec_db.device)
+        spec_db = torch.max(spec_db, new_spec_db_max)
+
     return spec_db
 
 
+@torch.jit.script
 def create_dct(n_mfcc, n_mels, norm):
-    # type: (int, int, string) -> Tensor
+    # type: (int, int, Optional[str]) -> Tensor
     """
     Creates a DCT transformation matrix with shape (num_mels, num_mfcc),
     normalized depending on norm
@@ -237,7 +219,7 @@ def create_dct(n_mfcc, n_mels, norm):
     Inputs:
         n_mfcc (int) : number of mfc coefficients to retain
         n_mels (int): number of MEL bins
-        norm (string) : norm to use
+        norm (Optional[str]) : norm to use (either 'ortho' or None)
 
     Outputs:
         Tensor: The transformation matrix, to be right-multiplied to row-wise data.
@@ -245,48 +227,19 @@ def create_dct(n_mfcc, n_mels, norm):
     outdim = n_mfcc
     dim = n_mels
     # http://en.wikipedia.org/wiki/Discrete_cosine_transform#DCT-II
-    n = torch.arange(dim, dtype=torch.get_default_dtype())
-    k = torch.arange(outdim, dtype=torch.get_default_dtype())[:, None]
-    dct = torch.cos(math.pi / dim * (n + 0.5) * k)
-    if norm == 'ortho':
-        dct[0] *= 1.0 / math.sqrt(2.0)
-        dct *= math.sqrt(2.0 / dim)
+    n = torch.arange(dim)
+    k = torch.arange(outdim)[:, None]
+    dct = torch.cos(math.pi / float(dim) * (n + 0.5) * k)
+    if norm is None:
+        dct *= 2.0
     else:
-        dct *= 2
+        assert norm == 'ortho'
+        dct[0] *= 1.0 / math.sqrt(2.0)
+        dct *= math.sqrt(2.0 / float(dim))
     return dct.t()
 
 
-def MFCC(sig, mel_spect, log_mels, s2db, dct_mat):
-    # type: (Tensor, MelSpectrogram, bool, SpectrogramToDB, Tensor) -> Tensor
-    """Create the Mel-frequency cepstrum coefficients from an audio signal
-
-    By default, this calculates the MFCC on the DB-scaled Mel spectrogram.
-    This is not the textbook implementation, but is implemented here to
-    give consistency with librosa.
-
-    This output depends on the maximum value in the input spectrogram, and so
-    may return different values for an audio clip split into snippets vs. a
-    a full clip.
-
-    Inputs:
-        sig (Tensor): Tensor of audio of size (channels [c], samples [n])
-        mel_spect (MelSpectrogram): melspectrogram of sig
-        log_mels (bool): whether to use log-mel spectrograms instead of db-scaled
-        s2db (SpectrogramToDB): a SpectrogramToDB instance
-        dct_mat (Tensor): The transformation matrix (dct matrix), to be
-            right-multiplied to row-wise data
-    Outputs:
-        Tensor: Mel-frequency cepstrum coefficients
-    """
-    if log_mels:
-        log_offset = 1e-6
-        mel_spect = torch.log(mel_spect + log_offset)
-    else:
-        mel_spect = s2db(mel_spect)
-    mfcc = torch.matmul(mel_spect, dct_mat.to(mel_spect.device))
-    return mfcc
-
-
+@torch.jit.script
 def BLC2CBL(tensor):
     # type: (Tensor) -> Tensor
     """Permute a 3d tensor from Bands x Sample length x Channels to Channels x
@@ -301,6 +254,7 @@ def BLC2CBL(tensor):
     return tensor.permute(2, 0, 1).contiguous()
 
 
+@torch.jit.script
 def mu_law_encoding(x, qc):
     # type: (Tensor, int) -> Tensor
     """Encode signal based on mu-law companding.  For more info see the
@@ -318,7 +272,7 @@ def mu_law_encoding(x, qc):
     """
     assert isinstance(x, torch.Tensor), 'mu_law_encoding expects a Tensor'
     mu = qc - 1.
-    if not x.dtype.is_floating_point:
+    if not x.is_floating_point():
         x = x.to(torch.float)
     mu = torch.tensor(mu, dtype=x.dtype)
     x_mu = torch.sign(x) * torch.log1p(mu *
@@ -327,6 +281,7 @@ def mu_law_encoding(x, qc):
     return x_mu
 
 
+@torch.jit.script
 def mu_law_expanding(x_mu, qc):
     # type: (Tensor, int) -> Tensor
     """Decode mu-law encoded signal.  For more info see the
@@ -344,7 +299,7 @@ def mu_law_expanding(x_mu, qc):
     """
     assert isinstance(x_mu, torch.Tensor), 'mu_law_expanding expects a Tensor'
     mu = qc - 1.
-    if not x_mu.dtype.is_floating_point:
+    if not x_mu.is_floating_point():
         x_mu = x_mu.to(torch.float)
     mu = torch.tensor(mu, dtype=x_mu.dtype)
     x = ((x_mu) / mu) * 2 - 1.

torchaudio/transforms.py

@@ -2,9 +2,11 @@ from __future__ import division, print_function
 from warnings import warn
 import math
 import torch
+from typing import Optional
 from . import functional as F
 
 
+# TODO remove this class
 class Compose(object):
     """Composes several transforms together.
 
@@ -17,7 +19,6 @@ class Compose(object):
         >>>     transforms.PadTrim(max_len=16000),
         >>> ])
     """
-
     def __init__(self, transforms):
         self.transforms = transforms
 
@@ -35,7 +36,7 @@ class Compose(object):
         return format_string
 
 
-class Scale(object):
+class Scale(torch.jit.ScriptModule):
     """Scale audio tensor from a 16-bit integer (represented as a FloatTensor)
     to a floating point number between -1.0 and 1.0.  Note the 16-bit number is
     called the "bit depth" or "precision", not to be confused with "bit rate".
@@ -44,11 +45,14 @@ class Scale(object):
         factor (int): maximum value of input tensor. default: 16-bit depth
 
     """
+    __constants__ = ['factor']
 
     def __init__(self, factor=2**31):
+        super(Scale, self).__init__()
         self.factor = factor
 
-    def __call__(self, tensor):
+    @torch.jit.script_method
+    def forward(self, tensor):
         """
 
         Args:
@@ -64,7 +68,7 @@ class Scale(object):
         return self.__class__.__name__ + '()'
 
 
-class PadTrim(object):
+class PadTrim(torch.jit.ScriptModule):
     """Pad/Trim a 2d-Tensor (Signal or Labels)
 
     Args:
@@ -73,13 +77,16 @@ class PadTrim(object):
         channels_first (bool): Pad for channels first tensors.  Default: `True`
 
     """
+    __constants__ = ['max_len', 'fill_value', 'len_dim', 'ch_dim']
 
-    def __init__(self, max_len, fill_value=0, channels_first=True):
+    def __init__(self, max_len, fill_value=0., channels_first=True):
+        super(PadTrim, self).__init__()
         self.max_len = max_len
         self.fill_value = fill_value
         self.len_dim, self.ch_dim = int(channels_first), int(not channels_first)
 
-    def __call__(self, tensor):
+    @torch.jit.script_method
+    def forward(self, tensor):
         """
 
         Returns:
@@ -92,7 +99,7 @@ class PadTrim(object):
         return self.__class__.__name__ + '(max_len={0})'.format(self.max_len)
 
 
-class DownmixMono(object):
+class DownmixMono(torch.jit.ScriptModule):
     """Downmix any stereo signals to mono.  Consider using a `SoxEffectsChain` with
        the `channels` effect instead of this transformation.
 
@@ -104,22 +111,29 @@ class DownmixMono(object):
         tensor (Tensor) (Samples x 1):
 
     """
+    __constants__ = ['ch_dim']
 
     def __init__(self, channels_first=None):
+        super(DownmixMono, self).__init__()
         self.ch_dim = int(not channels_first)
 
-    def __call__(self, tensor):
+    @torch.jit.script_method
+    def forward(self, tensor):
         return F.downmix_mono(tensor, self.ch_dim)
 
     def __repr__(self):
         return self.__class__.__name__ + '()'
 
 
-class LC2CL(object):
+class LC2CL(torch.jit.ScriptModule):
     """Permute a 2d tensor from samples (n x c) to (c x n)
     """
 
-    def __call__(self, tensor):
+    def __init__(self):
+        super(LC2CL, self).__init__()
+
+    @torch.jit.script_method
+    def forward(self, tensor):
         """
 
         Args:
@@ -139,7 +153,7 @@ def SPECTROGRAM(*args, **kwargs):
     return Spectrogram(*args, **kwargs)
 
 
-class Spectrogram(object):
+class Spectrogram(torch.jit.ScriptModule):
     """Create a spectrogram from a raw audio signal
 
     Args:
@@ -153,21 +167,25 @@ class Spectrogram(object):
         normalize (bool) : whether to normalize by magnitude after stft
         wkwargs (dict, optional): arguments for window function
     """
+    __constants__ = ['n_fft', 'ws', 'hop', 'pad', 'power', 'normalize']
+
     def __init__(self, n_fft=400, ws=None, hop=None,
                  pad=0, window=torch.hann_window,
                  power=2, normalize=False, wkwargs=None):
+        super(Spectrogram, self).__init__()
         self.n_fft = n_fft
         # number of fft bins. the returned STFT result will have n_fft // 2 + 1
         # number of frequecies due to onesided=True in torch.stft
         self.ws = ws if ws is not None else n_fft
         self.hop = hop if hop is not None else self.ws // 2
-        self.window = window(self.ws) if wkwargs is None else window(self.ws, **wkwargs)
+        window = window(self.ws) if wkwargs is None else window(self.ws, **wkwargs)
+        self.window = torch.jit.Attribute(window, torch.Tensor)
         self.pad = pad
         self.power = power
         self.normalize = normalize
-        self.wkwargs = wkwargs
 
-    def __call__(self, sig):
+    @torch.jit.script_method
+    def forward(self, sig):
         """
         Args:
             sig (Tensor): Tensor of audio of size (c, n)
@@ -188,10 +206,12 @@ def F2M(*args, **kwargs):
     return MelScale(*args, **kwargs)
 
 
-class MelScale(object):
+class MelScale(torch.jit.ScriptModule):
     """This turns a normal STFT into a mel frequency STFT, using a conversion
        matrix.  This uses triangular filter banks.
 
+       User can control which device the filter bank (`fb`) is (e.g. fb.to(spec_f.device)).
+
     Args:
         n_mels (int): number of mel bins
         sr (int): sample rate of audio signal
@@ -200,20 +220,30 @@ class MelScale(object):
         n_stft (int, optional): number of filter banks from stft. Calculated from first input
             if `None` is given.  See `n_fft` in `Spectrogram`.
     """
+    __constants__ = ['n_mels', 'sr', 'f_min', 'f_max']
+
     def __init__(self, n_mels=128, sr=16000, f_max=None, f_min=0., n_stft=None):
+        super(MelScale, self).__init__()
         self.n_mels = n_mels
         self.sr = sr
-        self.f_max = f_max if f_max is not None else sr // 2
+        self.f_max = f_max if f_max is not None else float(sr // 2)
         self.f_min = f_min
-        self.fb = F.create_fb_matrix(
-            n_stft, self.f_min, self.f_max, self.n_mels) if n_stft is not None else n_stft
-
-    def __call__(self, spec_f):
-        self.fb, spec_m = F.mel_scale(spec_f, self.f_min, self.f_max, self.n_mels, self.fb)
+        fb = torch.empty(0) if n_stft is None else F.create_fb_matrix(
+            n_stft, self.f_min, self.f_max, self.n_mels)
+        self.fb = torch.jit.Attribute(fb, torch.Tensor)
+
+    @torch.jit.script_method
+    def forward(self, spec_f):
+        if self.fb.numel() == 0:
+            tmp_fb = F.create_fb_matrix(spec_f.size(2), self.f_min, self.f_max, self.n_mels)
+            # Attributes cannot be reassigned outside __init__ so workaround
+            self.fb.resize_(tmp_fb.size())
+            self.fb.copy_(tmp_fb)
+        spec_m = torch.matmul(spec_f, self.fb)  # (c, l, n_fft) dot (n_fft, n_mels) -> (c, l, n_mels)
         return spec_m
 
 
-class SpectrogramToDB(object):
+class SpectrogramToDB(torch.jit.ScriptModule):
     """Turns a spectrogram from the power/amplitude scale to the decibel scale.
 
     This output depends on the maximum value in the input spectrogram, and so
@@ -226,23 +256,27 @@ class SpectrogramToDB(object):
         top_db (float, optional): minimum negative cut-off in decibels.  A reasonable number
             is 80.
     """
+    __constants__ = ['multiplier', 'amin', 'ref_value', 'db_multiplier']
+
     def __init__(self, stype="power", top_db=None):
-        self.stype = stype
+        super(SpectrogramToDB, self).__init__()
+        self.stype = torch.jit.Attribute(stype, str)
         if top_db is not None and top_db < 0:
             raise ValueError('top_db must be positive value')
-        self.top_db = top_db
+        self.top_db = torch.jit.Attribute(top_db, Optional[float])
         self.multiplier = 10. if stype == "power" else 20.
         self.amin = 1e-10
         self.ref_value = 1.
         self.db_multiplier = math.log10(max(self.amin, self.ref_value))
 
-    def __call__(self, spec):
+    @torch.jit.script_method
+    def forward(self, spec):
         # numerically stable implementation from librosa
         # https://librosa.github.io/librosa/_modules/librosa/core/spectrum.html
         return F.spectrogram_to_DB(spec, self.multiplier, self.amin, self.db_multiplier, self.top_db)
 
 
-class MFCC(object):
+class MFCC(torch.jit.ScriptModule):
     """Create the Mel-frequency cepstrum coefficients from an audio signal
 
         By default, this calculates the MFCC on the DB-scaled Mel spectrogram.
@@ -257,21 +291,22 @@ class MFCC(object):
         sr (int) : sample rate of audio signal
         n_mfcc (int) : number of mfc coefficients to retain
         dct_type (int) : type of DCT (discrete cosine transform) to use
-        norm (string) : norm to use
+        norm (string, optional) : norm to use
         log_mels (bool) : whether to use log-mel spectrograms instead of db-scaled
         melkwargs (dict, optional): arguments for MelSpectrogram
     """
+    __constants__ = ['sr', 'n_mfcc', 'dct_type', 'top_db', 'log_mels']
+
     def __init__(self, sr=16000, n_mfcc=40, dct_type=2, norm='ortho', log_mels=False,
                  melkwargs=None):
-
+        super(MFCC, self).__init__()
         supported_dct_types = [2]
         if dct_type not in supported_dct_types:
             raise ValueError('DCT type not supported'.format(dct_type))
         self.sr = sr
         self.n_mfcc = n_mfcc
         self.dct_type = dct_type
-        self.norm = norm
-        self.melkwargs = melkwargs
+        self.norm = torch.jit.Attribute(norm, Optional[str])
         self.top_db = 80.
         self.s2db = SpectrogramToDB("power", self.top_db)
 
@@ -282,10 +317,12 @@ class MFCC(object):
 
         if self.n_mfcc > self.MelSpectrogram.n_mels:
             raise ValueError('Cannot select more MFCC coefficients than # mel bins')
-        self.dct_mat = F.create_dct(self.n_mfcc, self.MelSpectrogram.n_mels, self.norm)
+        dct_mat = F.create_dct(self.n_mfcc, self.MelSpectrogram.n_mels, self.norm)
+        self.dct_mat = torch.jit.Attribute(dct_mat, torch.Tensor)
         self.log_mels = log_mels
 
-    def __call__(self, sig):
+    @torch.jit.script_method
+    def forward(self, sig):
         """
         Args:
             sig (Tensor): Tensor of audio of size (channels [c], samples [n])
@@ -295,10 +332,17 @@ class MFCC(object):
                 is unchanged, hops is the number of hops, and n_mels is the
                 number of mel bins.
         """
-        return F.MFCC(sig, self.MelSpectrogram(sig), self.log_mels, self.s2db, self.dct_mat)
+        mel_spect = self.MelSpectrogram(sig)
+        if self.log_mels:
+            log_offset = 1e-6
+            mel_spect = torch.log(mel_spect + log_offset)
+        else:
+            mel_spect = self.s2db(mel_spect)
+        mfcc = torch.matmul(mel_spect, self.dct_mat)
+        return mfcc
 
 
-class MelSpectrogram(object):
+class MelSpectrogram(torch.jit.ScriptModule):
     """Create MEL Spectrograms from a raw audio signal using the stft
        function in PyTorch.
 
@@ -323,27 +367,26 @@ class MelSpectrogram(object):
         >>> sig, sr = torchaudio.load("test.wav", normalization=True)
         >>> spec_mel = transforms.MelSpectrogram(sr)(sig)  # (c, l, m)
     """
+    __constants__ = ['sr', 'n_fft', 'ws', 'hop', 'pad', 'n_mels', 'f_min']
+
     def __init__(self, sr=16000, n_fft=400, ws=None, hop=None, f_min=0., f_max=None,
                  pad=0, n_mels=128, window=torch.hann_window, wkwargs=None):
-        self.window = window
+        super(MelSpectrogram, self).__init__()
         self.sr = sr
         self.n_fft = n_fft
         self.ws = ws if ws is not None else n_fft
         self.hop = hop if hop is not None else self.ws // 2
         self.pad = pad
         self.n_mels = n_mels  # number of mel frequency bins
-        self.wkwargs = wkwargs
-        self.f_max = f_max
+        self.f_max = torch.jit.Attribute(f_max, Optional[float])
         self.f_min = f_min
         self.spec = Spectrogram(n_fft=self.n_fft, ws=self.ws, hop=self.hop,
-                                pad=self.pad, window=self.window, power=2,
-                                normalize=False, wkwargs=self.wkwargs)
+                                pad=self.pad, window=window, power=2,
+                                normalize=False, wkwargs=wkwargs)
         self.fm = MelScale(self.n_mels, self.sr, self.f_max, self.f_min)
-        self.transforms = Compose([
-            self.spec, self.fm
-        ])
 
-    def __call__(self, sig):
+    @torch.jit.script_method
+    def forward(self, sig):
         """
         Args:
             sig (Tensor): Tensor of audio of size (channels [c], samples [n])
@@ -354,8 +397,8 @@ class MelSpectrogram(object):
                 number of mel bins.
 
         """
-        spec_mel = self.transforms(sig)
-
+        spec = self.spec(sig)
+        spec_mel = self.fm(spec)
         return spec_mel
 
 
@@ -363,12 +406,16 @@ def MEL(*args, **kwargs):
     raise DeprecationWarning("MEL has been removed from the library please use MelSpectrogram or librosa")
 
 
-class BLC2CBL(object):
+class BLC2CBL(torch.jit.ScriptModule):
     """Permute a 3d tensor from Bands x Sample length x Channels to Channels x
        Bands x Samples length
     """
 
-    def __call__(self, tensor):
+    def __init__(self):
+        super(BLC2CBL, self).__init__()
+
+    @torch.jit.script_method
+    def forward(self, tensor):
         """
 
         Args:
@@ -384,7 +431,7 @@ class BLC2CBL(object):
         return self.__class__.__name__ + '()'
 
 
-class MuLawEncoding(object):
+class MuLawEncoding(torch.jit.ScriptModule):
     """Encode signal based on mu-law companding.  For more info see the
     `Wikipedia Entry <https://en.wikipedia.org/wiki/%CE%9C-law_algorithm>`_
 
@@ -395,11 +442,14 @@ class MuLawEncoding(object):
         quantization_channels (int): Number of channels. default: 256
 
     """
+    __constants__ = ['qc']
 
     def __init__(self, quantization_channels=256):
+        super(MuLawEncoding, self).__init__()
         self.qc = quantization_channels
 
-    def __call__(self, x):
+    @torch.jit.script_method
+    def forward(self, x):
         """
 
         Args:
@@ -415,7 +465,7 @@ class MuLawEncoding(object):
         return self.__class__.__name__ + '()'
 
 
-class MuLawExpanding(object):
+class MuLawExpanding(torch.jit.ScriptModule):
     """Decode mu-law encoded signal.  For more info see the
     `Wikipedia Entry <https://en.wikipedia.org/wiki/%CE%9C-law_algorithm>`_
 
@@ -426,11 +476,14 @@ class MuLawExpanding(object):
         quantization_channels (int): Number of channels. default: 256
 
     """
+    __constants__ = ['qc']
 
     def __init__(self, quantization_channels=256):
+        super(MuLawExpanding, self).__init__()
         self.qc = quantization_channels
 
-    def __call__(self, x_mu):
+    @torch.jit.script_method
+    def forward(self, x_mu):
         """
 
         Args: