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
|
import sys
import math
import unittest
import torch
import torchaudio
from .. import common_utils
@common_utils.skipIfNoSoxBackend
class Test_SoxEffectsChain(common_utils.TorchaudioTestCase):
backend = 'sox'
test_filepath = common_utils.get_asset_path("steam-train-whistle-daniel_simon.wav")
def test_single_channel(self):
fn_sine = common_utils.get_asset_path("sinewave.wav")
E = torchaudio.sox_effects.SoxEffectsChain()
E.set_input_file(fn_sine)
E.append_effect_to_chain("echos", [0.8, 0.7, 40, 0.25, 63, 0.3])
x, sr = E.sox_build_flow_effects()
# check if effects worked
# print(x.size())
def test_rate_channels(self):
target_rate = 16000
target_channels = 1
E = torchaudio.sox_effects.SoxEffectsChain()
E.set_input_file(self.test_filepath)
E.append_effect_to_chain("rate", [target_rate])
E.append_effect_to_chain("channels", [target_channels])
x, sr = E.sox_build_flow_effects()
# check if effects worked
self.assertEqual(sr, target_rate)
self.assertEqual(x.size(0), target_channels)
@unittest.skipIf(sys.platform == 'darwin', 'This test is known to fail on macOS')
def test_lowpass_speed(self):
speed = .8
si, _ = torchaudio.info(self.test_filepath)
E = torchaudio.sox_effects.SoxEffectsChain()
E.set_input_file(self.test_filepath)
E.append_effect_to_chain("lowpass", 100)
E.append_effect_to_chain("speed", speed)
E.append_effect_to_chain("rate", si.rate)
x, sr = E.sox_build_flow_effects()
# check if effects worked, add small tolerance for rounding effects
self.assertEqual(x.size(1), int((si.length / si.channels) / speed), atol=1, rtol=1e-8)
def test_ulaw_and_siginfo(self):
si_out = torchaudio.sox_signalinfo_t()
ei_out = torchaudio.sox_encodinginfo_t()
si_out.precision = 8
ei_out.encoding = torchaudio.get_sox_encoding_t(9)
ei_out.bits_per_sample = 8
si_in, ei_in = torchaudio.info(self.test_filepath)
si_out.rate = 44100
si_out.channels = 2
E = torchaudio.sox_effects.SoxEffectsChain(out_siginfo=si_out, out_encinfo=ei_out)
E.set_input_file(self.test_filepath)
x, sr = E.sox_build_flow_effects()
# Note: the output was encoded into ulaw because the
# number of unique values in the output is less than 256.
self.assertLess(x.unique().size(0), 2**8 + 1)
self.assertEqual(x.numel(), si_in.length)
def test_band_chorus(self):
si_in, ei_in = torchaudio.info(self.test_filepath)
ei_in.encoding = torchaudio.get_sox_encoding_t(1)
E = torchaudio.sox_effects.SoxEffectsChain(out_encinfo=ei_in, out_siginfo=si_in)
E.set_input_file(self.test_filepath)
E.append_effect_to_chain("band", ["-n", "10k", "3.5k"])
E.append_effect_to_chain("chorus", [.5, .7, 55, 0.4, .25, 2, '-s'])
E.append_effect_to_chain("rate", [si_in.rate])
E.append_effect_to_chain("channels", [si_in.channels])
x, sr = E.sox_build_flow_effects()
# The chorus effect will make the output file longer than the input
self.assertEqual(x.size(0), si_in.channels)
self.assertGreaterEqual(x.size(1) * x.size(0), si_in.length)
def test_synth(self):
si_in, ei_in = torchaudio.info(self.test_filepath)
len_in_seconds = si_in.length / si_in.channels / si_in.rate
ei_in.encoding = torchaudio.get_sox_encoding_t(1)
E = torchaudio.sox_effects.SoxEffectsChain(out_encinfo=ei_in, out_siginfo=si_in)
E.set_input_file(self.test_filepath)
E.append_effect_to_chain("synth", [str(len_in_seconds), "pinknoise", "mix"])
E.append_effect_to_chain("rate", [44100])
E.append_effect_to_chain("channels", [2])
x, sr = E.sox_build_flow_effects()
self.assertEqual(x.size(0), si_in.channels)
self.assertEqual(si_in.length, x.size(0) * x.size(1))
def test_gain(self):
E = torchaudio.sox_effects.SoxEffectsChain()
E.set_input_file(self.test_filepath)
E.append_effect_to_chain("gain", ["5"])
x, sr = E.sox_build_flow_effects()
E.clear_chain()
self.assertTrue(x.abs().max().item(), 1.)
E.set_input_file(self.test_filepath)
E.append_effect_to_chain("gain", ["-e", "-5"])
x, sr = E.sox_build_flow_effects()
E.clear_chain()
self.assertLess(x.abs().max().item(), 1.)
E.set_input_file(self.test_filepath)
E.append_effect_to_chain("gain", ["-b", "8"])
x, sr = E.sox_build_flow_effects()
E.clear_chain()
self.assertTrue(x.abs().max().item(), 1.)
E.set_input_file(self.test_filepath)
E.append_effect_to_chain("gain", ["-n", "-10"])
x, sr = E.sox_build_flow_effects()
E.clear_chain()
self.assertLess(x.abs().max().item(), 1.)
def test_tempo_or_speed(self):
tempo = .8
si, _ = torchaudio.info(self.test_filepath)
E = torchaudio.sox_effects.SoxEffectsChain()
E.set_input_file(self.test_filepath)
E.append_effect_to_chain("tempo", ["-s", tempo])
x, sr = E.sox_build_flow_effects()
# check if effect worked
self.assertAlmostEqual(x.size(1), math.ceil((si.length / si.channels) / tempo), delta=1)
# tempo > 1
E.clear_chain()
tempo = 1.2
E.append_effect_to_chain("tempo", ["-s", tempo])
x, sr = E.sox_build_flow_effects()
# check if effect worked
self.assertAlmostEqual(x.size(1), math.ceil((si.length / si.channels) / tempo), delta=1)
# tempo > 1
E.clear_chain()
speed = 1.2
E.append_effect_to_chain("speed", [speed])
E.append_effect_to_chain("rate", [si.rate])
x, sr = E.sox_build_flow_effects()
# check if effect worked
self.assertAlmostEqual(x.size(1), math.ceil((si.length / si.channels) / speed), delta=1)
# speed < 1
E.clear_chain()
speed = 0.8
E.append_effect_to_chain("speed", [speed])
E.append_effect_to_chain("rate", [si.rate])
x, sr = E.sox_build_flow_effects()
# check if effect worked
self.assertAlmostEqual(x.size(1), math.ceil((si.length / si.channels) / speed), delta=1)
def test_trim(self):
x_orig, _ = torchaudio.load(self.test_filepath)
offset = "10000s"
offset_int = int(offset[:-1])
num_frames = "20000s"
num_frames_int = int(num_frames[:-1])
E = torchaudio.sox_effects.SoxEffectsChain()
E.set_input_file(self.test_filepath)
E.append_effect_to_chain("trim", [offset, num_frames])
x, sr = E.sox_build_flow_effects()
# check if effect worked
self.assertTrue(x.allclose(x_orig[:, offset_int:(offset_int + num_frames_int)], rtol=1e-4, atol=1e-4))
def test_silence_contrast(self):
si, _ = torchaudio.info(self.test_filepath)
E = torchaudio.sox_effects.SoxEffectsChain()
E.set_input_file(self.test_filepath)
E.append_effect_to_chain("silence", [1, 100, 1])
E.append_effect_to_chain("contrast", [])
x, sr = E.sox_build_flow_effects()
# check if effect worked
self.assertLess(x.numel(), si.length)
def test_reverse(self):
x_orig, _ = torchaudio.load(self.test_filepath)
E = torchaudio.sox_effects.SoxEffectsChain()
E.set_input_file(self.test_filepath)
E.append_effect_to_chain("reverse", "")
x_rev, _ = E.sox_build_flow_effects()
# check if effect worked
rev_idx = torch.LongTensor(range(x_orig.size(1))[::-1])
self.assertTrue(x_orig.allclose(x_rev[:, rev_idx], rtol=1e-5, atol=2e-5))
def test_compand_fade(self):
E = torchaudio.sox_effects.SoxEffectsChain()
E.set_input_file(self.test_filepath)
E.append_effect_to_chain("compand", ["0.3,1", "6:-70,-60,-20", "-5", "-90", "0.2"])
E.append_effect_to_chain("fade", ["q", "0.25", "0", "0.33"])
x, _ = E.sox_build_flow_effects()
# check if effect worked
# print(x.size())
def test_biquad_delay(self):
si, _ = torchaudio.info(self.test_filepath)
E = torchaudio.sox_effects.SoxEffectsChain()
E.set_input_file(self.test_filepath)
E.append_effect_to_chain("biquad", ["0.25136437", "0.50272873", "0.25136437",
"1.0", "-0.17123075", "0.17668821"])
E.append_effect_to_chain("delay", ["15000s"])
x, _ = E.sox_build_flow_effects()
# check if effect worked
self.assertTrue(x.size(1) == (si.length / si.channels) + 15000)
def test_invalid_effect_name(self):
E = torchaudio.sox_effects.SoxEffectsChain()
E.set_input_file(self.test_filepath)
# there is no effect named "special"
with self.assertRaises(LookupError):
E.append_effect_to_chain("special", [""])
def test_unimplemented_effect(self):
E = torchaudio.sox_effects.SoxEffectsChain()
E.set_input_file(self.test_filepath)
# the sox spectrogram function is not implemented in torchaudio
with self.assertRaises(NotImplementedError):
E.append_effect_to_chain("spectrogram", [""])
def test_invalid_effect_options(self):
E = torchaudio.sox_effects.SoxEffectsChain()
E.set_input_file(self.test_filepath)
# first two options should be combined to "0.3,1"
E.append_effect_to_chain("compand", ["0.3", "1", "6:-70,-60,-20", "-5", "-90", "0.2"])
with self.assertRaises(RuntimeError):
E.sox_build_flow_effects()
|
