File size: 10,876 Bytes
46a75d7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
import unittest

import torch as T

from TTS.tts.layers.losses import BCELossMasked, L1LossMasked, MSELossMasked, SSIMLoss
from TTS.tts.utils.helpers import sequence_mask


class L1LossMaskedTests(unittest.TestCase):
    def test_in_out(self):  # pylint: disable=no-self-use
        # test input == target
        layer = L1LossMasked(seq_len_norm=False)
        dummy_input = T.ones(4, 8, 128).float()
        dummy_target = T.ones(4, 8, 128).float()
        dummy_length = (T.ones(4) * 8).long()
        output = layer(dummy_input, dummy_target, dummy_length)
        assert output.item() == 0.0

        # test input != target
        dummy_input = T.ones(4, 8, 128).float()
        dummy_target = T.zeros(4, 8, 128).float()
        dummy_length = (T.ones(4) * 8).long()
        output = layer(dummy_input, dummy_target, dummy_length)
        assert output.item() == 1.0, "1.0 vs {}".format(output.item())

        # test if padded values of input makes any difference
        dummy_input = T.ones(4, 8, 128).float()
        dummy_target = T.zeros(4, 8, 128).float()
        dummy_length = (T.arange(5, 9)).long()
        mask = ((sequence_mask(dummy_length).float() - 1.0) * 100.0).unsqueeze(2)
        output = layer(dummy_input + mask, dummy_target, dummy_length)
        assert output.item() == 1.0, "1.0 vs {}".format(output.item())

        dummy_input = T.rand(4, 8, 128).float()
        dummy_target = dummy_input.detach()
        dummy_length = (T.arange(5, 9)).long()
        mask = ((sequence_mask(dummy_length).float() - 1.0) * 100.0).unsqueeze(2)
        output = layer(dummy_input + mask, dummy_target, dummy_length)
        assert output.item() == 0, "0 vs {}".format(output.item())

        # seq_len_norm = True
        # test input == target
        layer = L1LossMasked(seq_len_norm=True)
        dummy_input = T.ones(4, 8, 128).float()
        dummy_target = T.ones(4, 8, 128).float()
        dummy_length = (T.ones(4) * 8).long()
        output = layer(dummy_input, dummy_target, dummy_length)
        assert output.item() == 0.0

        # test input != target
        dummy_input = T.ones(4, 8, 128).float()
        dummy_target = T.zeros(4, 8, 128).float()
        dummy_length = (T.ones(4) * 8).long()
        output = layer(dummy_input, dummy_target, dummy_length)
        assert output.item() == 1.0, "1.0 vs {}".format(output.item())

        # test if padded values of input makes any difference
        dummy_input = T.ones(4, 8, 128).float()
        dummy_target = T.zeros(4, 8, 128).float()
        dummy_length = (T.arange(5, 9)).long()
        mask = ((sequence_mask(dummy_length).float() - 1.0) * 100.0).unsqueeze(2)
        output = layer(dummy_input + mask, dummy_target, dummy_length)
        assert abs(output.item() - 1.0) < 1e-5, "1.0 vs {}".format(output.item())

        dummy_input = T.rand(4, 8, 128).float()
        dummy_target = dummy_input.detach()
        dummy_length = (T.arange(5, 9)).long()
        mask = ((sequence_mask(dummy_length).float() - 1.0) * 100.0).unsqueeze(2)
        output = layer(dummy_input + mask, dummy_target, dummy_length)
        assert output.item() == 0, "0 vs {}".format(output.item())


class MSELossMaskedTests(unittest.TestCase):
    def test_in_out(self):  # pylint: disable=no-self-use
        # test input == target
        layer = MSELossMasked(seq_len_norm=False)
        dummy_input = T.ones(4, 8, 128).float()
        dummy_target = T.ones(4, 8, 128).float()
        dummy_length = (T.ones(4) * 8).long()
        output = layer(dummy_input, dummy_target, dummy_length)
        assert output.item() == 0.0

        # test input != target
        dummy_input = T.ones(4, 8, 128).float()
        dummy_target = T.zeros(4, 8, 128).float()
        dummy_length = (T.ones(4) * 8).long()
        output = layer(dummy_input, dummy_target, dummy_length)
        assert output.item() == 1.0, "1.0 vs {}".format(output.item())

        # test if padded values of input makes any difference
        dummy_input = T.ones(4, 8, 128).float()
        dummy_target = T.zeros(4, 8, 128).float()
        dummy_length = (T.arange(5, 9)).long()
        mask = ((sequence_mask(dummy_length).float() - 1.0) * 100.0).unsqueeze(2)
        output = layer(dummy_input + mask, dummy_target, dummy_length)
        assert output.item() == 1.0, "1.0 vs {}".format(output.item())

        dummy_input = T.rand(4, 8, 128).float()
        dummy_target = dummy_input.detach()
        dummy_length = (T.arange(5, 9)).long()
        mask = ((sequence_mask(dummy_length).float() - 1.0) * 100.0).unsqueeze(2)
        output = layer(dummy_input + mask, dummy_target, dummy_length)
        assert output.item() == 0, "0 vs {}".format(output.item())

        # seq_len_norm = True
        # test input == target
        layer = MSELossMasked(seq_len_norm=True)
        dummy_input = T.ones(4, 8, 128).float()
        dummy_target = T.ones(4, 8, 128).float()
        dummy_length = (T.ones(4) * 8).long()
        output = layer(dummy_input, dummy_target, dummy_length)
        assert output.item() == 0.0

        # test input != target
        dummy_input = T.ones(4, 8, 128).float()
        dummy_target = T.zeros(4, 8, 128).float()
        dummy_length = (T.ones(4) * 8).long()
        output = layer(dummy_input, dummy_target, dummy_length)
        assert output.item() == 1.0, "1.0 vs {}".format(output.item())

        # test if padded values of input makes any difference
        dummy_input = T.ones(4, 8, 128).float()
        dummy_target = T.zeros(4, 8, 128).float()
        dummy_length = (T.arange(5, 9)).long()
        mask = ((sequence_mask(dummy_length).float() - 1.0) * 100.0).unsqueeze(2)
        output = layer(dummy_input + mask, dummy_target, dummy_length)
        assert abs(output.item() - 1.0) < 1e-5, "1.0 vs {}".format(output.item())

        dummy_input = T.rand(4, 8, 128).float()
        dummy_target = dummy_input.detach()
        dummy_length = (T.arange(5, 9)).long()
        mask = ((sequence_mask(dummy_length).float() - 1.0) * 100.0).unsqueeze(2)
        output = layer(dummy_input + mask, dummy_target, dummy_length)
        assert output.item() == 0, "0 vs {}".format(output.item())


class SSIMLossTests(unittest.TestCase):
    def test_in_out(self):  # pylint: disable=no-self-use
        # test input == target
        layer = SSIMLoss()
        dummy_input = T.ones(4, 57, 128).float()
        dummy_target = T.ones(4, 57, 128).float()
        dummy_length = (T.ones(4) * 8).long()
        output = layer(dummy_input, dummy_target, dummy_length)
        assert output.item() == 0.0

        # test input != target
        dummy_input = T.arange(0, 4 * 57 * 128)
        dummy_input = dummy_input.reshape(4, 57, 128).float()
        dummy_target = T.arange(-4 * 57 * 128, 0)
        dummy_target = dummy_target.reshape(4, 57, 128).float()
        dummy_target = -dummy_target

        dummy_length = (T.ones(4) * 58).long()
        output = layer(dummy_input, dummy_target, dummy_length)
        assert output.item() >= 1.0, "0 vs {}".format(output.item())

        # test if padded values of input makes any difference
        dummy_input = T.ones(4, 57, 128).float()
        dummy_target = T.zeros(4, 57, 128).float()
        dummy_length = (T.arange(54, 58)).long()
        mask = ((sequence_mask(dummy_length).float() - 1.0) * 100.0).unsqueeze(2)
        output = layer(dummy_input + mask, dummy_target, dummy_length)
        assert output.item() == 0.0

        dummy_input = T.rand(4, 57, 128).float()
        dummy_target = dummy_input.detach()
        dummy_length = (T.arange(54, 58)).long()
        mask = ((sequence_mask(dummy_length).float() - 1.0) * 100.0).unsqueeze(2)
        output = layer(dummy_input + mask, dummy_target, dummy_length)
        assert output.item() == 0, "0 vs {}".format(output.item())

        # seq_len_norm = True
        # test input == target
        layer = L1LossMasked(seq_len_norm=True)
        dummy_input = T.ones(4, 57, 128).float()
        dummy_target = T.ones(4, 57, 128).float()
        dummy_length = (T.ones(4) * 8).long()
        output = layer(dummy_input, dummy_target, dummy_length)
        assert output.item() == 0.0

        # test input != target
        dummy_input = T.ones(4, 57, 128).float()
        dummy_target = T.zeros(4, 57, 128).float()
        dummy_length = (T.ones(4) * 8).long()
        output = layer(dummy_input, dummy_target, dummy_length)
        assert output.item() == 1.0, "1.0 vs {}".format(output.item())

        # test if padded values of input makes any difference
        dummy_input = T.ones(4, 57, 128).float()
        dummy_target = T.zeros(4, 57, 128).float()
        dummy_length = (T.arange(54, 58)).long()
        mask = ((sequence_mask(dummy_length).float() - 1.0) * 100.0).unsqueeze(2)
        output = layer(dummy_input + mask, dummy_target, dummy_length)
        assert abs(output.item() - 1.0) < 1e-5, "1.0 vs {}".format(output.item())

        dummy_input = T.rand(4, 57, 128).float()
        dummy_target = dummy_input.detach()
        dummy_length = (T.arange(54, 58)).long()
        mask = ((sequence_mask(dummy_length).float() - 1.0) * 100.0).unsqueeze(2)
        output = layer(dummy_input + mask, dummy_target, dummy_length)
        assert output.item() == 0, "0 vs {}".format(output.item())


class BCELossTest(unittest.TestCase):
    def test_in_out(self):  # pylint: disable=no-self-use
        layer = BCELossMasked(pos_weight=5.0)

        length = T.tensor([95])
        target = (
            1.0 - sequence_mask(length - 1, 100).float()
        )  # [0, 0, .... 1, 1] where the first 1 is the last mel frame
        true_x = target * 200 - 100  # creates logits of [-100, -100, ... 100, 100] corresponding to target
        zero_x = T.zeros(target.shape) - 100.0  # simulate logits if it never stops decoding
        early_x = -200.0 * sequence_mask(length - 3, 100).float() + 100.0  # simulate logits on early stopping
        late_x = -200.0 * sequence_mask(length + 1, 100).float() + 100.0  # simulate logits on late stopping

        loss = layer(true_x, target, length)
        self.assertEqual(loss.item(), 0.0)

        loss = layer(early_x, target, length)
        self.assertAlmostEqual(loss.item(), 2.1053, places=4)

        loss = layer(late_x, target, length)
        self.assertAlmostEqual(loss.item(), 5.2632, places=4)

        loss = layer(zero_x, target, length)
        self.assertAlmostEqual(loss.item(), 5.2632, places=4)

        # pos_weight should be < 1 to penalize early stopping
        layer = BCELossMasked(pos_weight=0.2)
        loss = layer(true_x, target, length)
        self.assertEqual(loss.item(), 0.0)

        # when pos_weight < 1 overweight the early stopping loss

        loss_early = layer(early_x, target, length)
        loss_late = layer(late_x, target, length)
        self.assertGreater(loss_early.item(), loss_late.item())