Spaces:
Running
Running
File size: 6,775 Bytes
67c46fd |
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 |
"""SpecAugment module."""
from typing import Optional
from typing import Sequence
from typing import Union
from funasr_detach.models.specaug.mask_along_axis import MaskAlongAxis
from funasr_detach.models.specaug.mask_along_axis import MaskAlongAxisVariableMaxWidth
from funasr_detach.models.specaug.mask_along_axis import MaskAlongAxisLFR
from funasr_detach.models.specaug.time_warp import TimeWarp
from funasr_detach.register import tables
import torch.nn as nn
@tables.register("specaug_classes", "SpecAug")
class SpecAug(nn.Module):
"""Implementation of SpecAug.
Reference:
Daniel S. Park et al.
"SpecAugment: A Simple Data
Augmentation Method for Automatic Speech Recognition"
.. warning::
When using cuda mode, time_warp doesn't have reproducibility
due to `torch.nn.functional.interpolate`.
"""
def __init__(
self,
apply_time_warp: bool = True,
time_warp_window: int = 5,
time_warp_mode: str = "bicubic",
apply_freq_mask: bool = True,
freq_mask_width_range: Union[int, Sequence[int]] = (0, 20),
num_freq_mask: int = 2,
apply_time_mask: bool = True,
time_mask_width_range: Optional[Union[int, Sequence[int]]] = None,
time_mask_width_ratio_range: Optional[Union[float, Sequence[float]]] = None,
num_time_mask: int = 2,
):
if not apply_time_warp and not apply_time_mask and not apply_freq_mask:
raise ValueError(
"Either one of time_warp, time_mask, or freq_mask should be applied"
)
if (
apply_time_mask
and (time_mask_width_range is not None)
and (time_mask_width_ratio_range is not None)
):
raise ValueError(
'Either one of "time_mask_width_range" or '
'"time_mask_width_ratio_range" can be used'
)
super().__init__()
self.apply_time_warp = apply_time_warp
self.apply_freq_mask = apply_freq_mask
self.apply_time_mask = apply_time_mask
if apply_time_warp:
self.time_warp = TimeWarp(window=time_warp_window, mode=time_warp_mode)
else:
self.time_warp = None
if apply_freq_mask:
self.freq_mask = MaskAlongAxis(
dim="freq",
mask_width_range=freq_mask_width_range,
num_mask=num_freq_mask,
)
else:
self.freq_mask = None
if apply_time_mask:
if time_mask_width_range is not None:
self.time_mask = MaskAlongAxis(
dim="time",
mask_width_range=time_mask_width_range,
num_mask=num_time_mask,
)
elif time_mask_width_ratio_range is not None:
self.time_mask = MaskAlongAxisVariableMaxWidth(
dim="time",
mask_width_ratio_range=time_mask_width_ratio_range,
num_mask=num_time_mask,
)
else:
raise ValueError(
'Either one of "time_mask_width_range" or '
'"time_mask_width_ratio_range" should be used.'
)
else:
self.time_mask = None
def forward(self, x, x_lengths=None):
if self.time_warp is not None:
x, x_lengths = self.time_warp(x, x_lengths)
if self.freq_mask is not None:
x, x_lengths = self.freq_mask(x, x_lengths)
if self.time_mask is not None:
x, x_lengths = self.time_mask(x, x_lengths)
return x, x_lengths
@tables.register("specaug_classes", "SpecAugLFR")
class SpecAugLFR(nn.Module):
"""Implementation of SpecAug.
lfr_rate：low frame rate
"""
def __init__(
self,
apply_time_warp: bool = True,
time_warp_window: int = 5,
time_warp_mode: str = "bicubic",
apply_freq_mask: bool = True,
freq_mask_width_range: Union[int, Sequence[int]] = (0, 20),
num_freq_mask: int = 2,
lfr_rate: int = 0,
apply_time_mask: bool = True,
time_mask_width_range: Optional[Union[int, Sequence[int]]] = None,
time_mask_width_ratio_range: Optional[Union[float, Sequence[float]]] = None,
num_time_mask: int = 2,
):
if not apply_time_warp and not apply_time_mask and not apply_freq_mask:
raise ValueError(
"Either one of time_warp, time_mask, or freq_mask should be applied"
)
if (
apply_time_mask
and (time_mask_width_range is not None)
and (time_mask_width_ratio_range is not None)
):
raise ValueError(
'Either one of "time_mask_width_range" or '
'"time_mask_width_ratio_range" can be used'
)
super().__init__()
self.apply_time_warp = apply_time_warp
self.apply_freq_mask = apply_freq_mask
self.apply_time_mask = apply_time_mask
if apply_time_warp:
self.time_warp = TimeWarp(window=time_warp_window, mode=time_warp_mode)
else:
self.time_warp = None
if apply_freq_mask:
self.freq_mask = MaskAlongAxisLFR(
dim="freq",
mask_width_range=freq_mask_width_range,
num_mask=num_freq_mask,
lfr_rate=lfr_rate + 1,
)
else:
self.freq_mask = None
if apply_time_mask:
if time_mask_width_range is not None:
self.time_mask = MaskAlongAxisLFR(
dim="time",
mask_width_range=time_mask_width_range,
num_mask=num_time_mask,
lfr_rate=lfr_rate + 1,
)
elif time_mask_width_ratio_range is not None:
self.time_mask = MaskAlongAxisVariableMaxWidth(
dim="time",
mask_width_ratio_range=time_mask_width_ratio_range,
num_mask=num_time_mask,
)
else:
raise ValueError(
'Either one of "time_mask_width_range" or '
'"time_mask_width_ratio_range" should be used.'
)
else:
self.time_mask = None
def forward(self, x, x_lengths=None):
if self.time_warp is not None:
x, x_lengths = self.time_warp(x, x_lengths)
if self.freq_mask is not None:
x, x_lengths = self.freq_mask(x, x_lengths)
if self.time_mask is not None:
x, x_lengths = self.time_mask(x, x_lengths)
return x, x_lengths
|