Add model Wav2Letter

docs/source/models.rst

@@ -0,0 +1,17 @@
+.. role:: hidden
+    :class: hidden-section
+
+torchaudio.models
+======================
+
+.. currentmodule:: torchaudio.models
+
+The models subpackage contains definitions of models for addressing common audio tasks.
+
+
+:hidden:`Wav2Letter`
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: Wav2Letter
+
+  .. automethod:: forward

test/test_models.py

@@ -0,0 +1,22 @@
+from __future__ import absolute_import, division, print_function, unicode_literals
+import unittest
+
+import torch
+from torchaudio.models import wav2letter
+
+
+class ModelTester(unittest.TestCase):
+    def test_wav2letter(self):
+        batch_size = 2
+        n_features = 1
+        input_length = 800
+
+        model = wav2letter()
+        x = torch.rand(batch_size, n_features, input_length)
+        out = model(x)
+
+        assert out.size() == (1, batch_size, 40)
+
+
+if __name__ == '__main__':
+    unittest.main()

torchaudio/models/__init__.py

@@ -0,0 +1 @@
+from .wav2letter import *

torchaudio/models/wav2letter.py

@@ -0,0 +1,79 @@
+import torch
+from torch import nn
+
+__all__ = ["Wav2Letter", "wav2letter"]
+
+
+class Wav2Letter(nn.Module):
+    r"""Wav2Letter model architecture from the `"Wav2Letter: an End-to-End ConvNet-based Speech Recognition System"
+     <https://arxiv.org/abs/1609.03193>`_ paper.
+
+    Args:
+        num_classes (int, optional): Number of classes to be classified. (Default: ``40``)
+        version (str, optional): Wav2Letter can use as input: ``waveform``, ```power_spectrum``
+         or ```mfcc``. (Default: ``waveform``)
+        n_input_features (int, optional): If is used ``power_spectrum`` or ```mfcc`` must be
+        specified the number of features that were extracted. (Default: ``None``)
+    """
+
+    def __init__(self, num_classes=40, version="waveform", n_input_features=None):
+        super(Wav2Letter, self).__init__()
+        n_input_features = 250 if not n_input_features else n_input_features
+
+        acoustic_model = nn.Sequential(
+            nn.Conv1d(in_channels=n_input_features, out_channels=250, kernel_size=48, stride=2, padding=23),
+            nn.ReLU(inplace=True),
+            nn.Conv1d(in_channels=250, out_channels=250, kernel_size=7, stride=1, padding=3),
+            nn.ReLU(inplace=True),
+            nn.Conv1d(in_channels=250, out_channels=250, kernel_size=7, stride=1, padding=3),
+            nn.ReLU(inplace=True),
+            nn.Conv1d(in_channels=250, out_channels=250, kernel_size=7, stride=1, padding=3),
+            nn.ReLU(inplace=True),
+            nn.Conv1d(in_channels=250, out_channels=250, kernel_size=7, stride=1, padding=3),
+            nn.ReLU(inplace=True),
+            nn.Conv1d(in_channels=250, out_channels=250, kernel_size=7, stride=1, padding=3),
+            nn.ReLU(inplace=True),
+            nn.Conv1d(in_channels=250, out_channels=250, kernel_size=7, stride=1, padding=3),
+            nn.ReLU(inplace=True),
+            nn.Conv1d(in_channels=250, out_channels=250, kernel_size=7, stride=1, padding=3),
+            nn.ReLU(inplace=True),
+            nn.Conv1d(in_channels=250, out_channels=2000, kernel_size=32, stride=1, padding=15),
+            nn.ReLU(inplace=True),
+            nn.Conv1d(in_channels=2000, out_channels=2000, kernel_size=1, stride=1, padding=0),
+            nn.ReLU(inplace=True),
+            nn.Conv1d(in_channels=2000, out_channels=num_classes, kernel_size=1, stride=1, padding=0),
+            nn.ReLU(inplace=True)
+        )
+
+        if version == "waveform":
+            waveform_model = nn.Sequential(
+                nn.Conv1d(in_channels=1, out_channels=250, kernel_size=250, stride=160, padding=45),
+                nn.ReLU(inplace=True)
+            )
+            self.acoustic_model = nn.Sequential(waveform_model, acoustic_model)
+
+        if version in ["power_spectrum", "mfcc"]:
+            self.acoustic_model = acoustic_model
+
+    def forward(self, x):
+        # type: (Tensor) -> Tensor
+        r"""
+        Args:
+            x (torch.Tensor): Tensor of dimension (batch_size, n_features, input_length).
+
+        Returns:
+            torch.Tensor: Predictor tensor of dimension (input_length, batch_size, number_of_classes).
+        """
+
+        x = self.acoustic_model(x)
+        x = nn.functional.log_softmax(x, dim=1)
+        x = x.permute(2, 0, 1)
+        return x
+
+
+def wav2letter(**kwargs):
+    r"""Wav2Letter model architecture from the `"Wav2Letter: an End-to-End ConvNet-based Speech Recognition System"
+     <https://arxiv.org/abs/1609.03193>`_ paper.
+    """
+    model = Wav2Letter(**kwargs)
+    return model