andybi7676
commited on
Commit
•
467bbf1
1
Parent(s):
1181389
end-2-end reborn model for mls-dutch unsupervised phoneme recognition (iter2-stage1)
Browse files- config.json +83 -0
- configuration_reborn.py +105 -0
- modeling_reborn.py +381 -0
- pytorch_model.bin +3 -0
config.json
ADDED
@@ -0,0 +1,83 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"architectures": [
|
3 |
+
"RebornUASRModel"
|
4 |
+
],
|
5 |
+
"auto_map": {
|
6 |
+
"AutoConfig": "configuration_reborn.RebornUASRConfig",
|
7 |
+
"AutoModel": "modeling_reborn.RebornUASRModel"
|
8 |
+
},
|
9 |
+
"discriminator_act_after_linear": false,
|
10 |
+
"discriminator_causal": true,
|
11 |
+
"discriminator_depth": 1,
|
12 |
+
"discriminator_dilation": 1,
|
13 |
+
"discriminator_dim": 256,
|
14 |
+
"discriminator_dropout": 0.0,
|
15 |
+
"discriminator_input_dim": 46,
|
16 |
+
"discriminator_kernel": 3,
|
17 |
+
"discriminator_linear_emb": false,
|
18 |
+
"discriminator_max_pool": false,
|
19 |
+
"discriminator_spectral_norm": false,
|
20 |
+
"discriminator_weight_norm": false,
|
21 |
+
"generator_bias": false,
|
22 |
+
"generator_bn_apply": false,
|
23 |
+
"generator_bn_init_weight": 30.0,
|
24 |
+
"generator_dilation": 1,
|
25 |
+
"generator_dropout": 0.0,
|
26 |
+
"generator_input_dim": 512,
|
27 |
+
"generator_kernel": 4,
|
28 |
+
"generator_output_dim": 46,
|
29 |
+
"generator_stride": 1,
|
30 |
+
"model_type": "reborn_uasr",
|
31 |
+
"phones": [
|
32 |
+
"\u0259",
|
33 |
+
"r",
|
34 |
+
"n",
|
35 |
+
"t",
|
36 |
+
"s",
|
37 |
+
"l",
|
38 |
+
"k",
|
39 |
+
"a\u02d0",
|
40 |
+
"d",
|
41 |
+
"\u0251",
|
42 |
+
"\u025b",
|
43 |
+
"i",
|
44 |
+
"\u026a",
|
45 |
+
"m",
|
46 |
+
"\u0263",
|
47 |
+
"\u0254",
|
48 |
+
"v",
|
49 |
+
"p",
|
50 |
+
"o\u02d0",
|
51 |
+
"b",
|
52 |
+
"e\u02d0",
|
53 |
+
"x",
|
54 |
+
"h",
|
55 |
+
"z",
|
56 |
+
"\u028b",
|
57 |
+
"\u025b\u026a",
|
58 |
+
"\u014b",
|
59 |
+
"f",
|
60 |
+
"\u0275",
|
61 |
+
"u",
|
62 |
+
"j",
|
63 |
+
"y",
|
64 |
+
"\u0153y",
|
65 |
+
"\u026a\u02d0",
|
66 |
+
"\u028c\u028a",
|
67 |
+
"\u0254\u02d0",
|
68 |
+
"\u00f8\u02d0",
|
69 |
+
"\u0283",
|
70 |
+
"t\u02b2",
|
71 |
+
"\u027e",
|
72 |
+
"w",
|
73 |
+
"<SIL>"
|
74 |
+
],
|
75 |
+
"segmenter_dropout": 0.1,
|
76 |
+
"segmenter_hidden_dim": 512,
|
77 |
+
"segmenter_input_dim": 512,
|
78 |
+
"segmenter_kernel_size": 7,
|
79 |
+
"segmenter_type": "cnn",
|
80 |
+
"special_token_nums": 4,
|
81 |
+
"torch_dtype": "float32",
|
82 |
+
"transformers_version": "4.24.0"
|
83 |
+
}
|
configuration_reborn.py
ADDED
@@ -0,0 +1,105 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import json
|
3 |
+
from transformers import PretrainedConfig
|
4 |
+
|
5 |
+
class RebornUASRConfig(PretrainedConfig):
|
6 |
+
'''
|
7 |
+
We can use this class to define the configuration of the reborn model.
|
8 |
+
The reborn UASR is composed of a segmenter, a discriminator, and a generator.
|
9 |
+
We only include the required configurations for the discriminator and the generator from fairseq's wav2vec-U model configuration.
|
10 |
+
'''
|
11 |
+
model_type = "reborn_uasr"
|
12 |
+
|
13 |
+
def __init__(self,
|
14 |
+
segmenter_type: str = "cnn",
|
15 |
+
segmenter_input_dim: int = 512,
|
16 |
+
segmenter_hidden_dim: int = 512,
|
17 |
+
segmenter_dropout: float = 0.1,
|
18 |
+
segmenter_kernel_size: int = 7,
|
19 |
+
|
20 |
+
discriminator_input_dim: int = 512,
|
21 |
+
discriminator_kernel: int = 3,
|
22 |
+
discriminator_dilation: int = 1,
|
23 |
+
discriminator_dim: int = 256,
|
24 |
+
discriminator_causal: bool = True,
|
25 |
+
discriminator_linear_emb: bool = False,
|
26 |
+
discriminator_depth: int = 1,
|
27 |
+
discriminator_max_pool: bool = False,
|
28 |
+
discriminator_act_after_linear: bool = False,
|
29 |
+
discriminator_dropout: float = 0.0,
|
30 |
+
discriminator_spectral_norm: bool = False,
|
31 |
+
discriminator_weight_norm: bool = False,
|
32 |
+
|
33 |
+
generator_input_dim: int = 512,
|
34 |
+
generator_output_dim: int = 40,
|
35 |
+
generator_kernel: int = 4,
|
36 |
+
generator_dilation: int = 1,
|
37 |
+
generator_stride: int = 1,
|
38 |
+
generator_bias: bool = False,
|
39 |
+
generator_dropout: float = 0.0,
|
40 |
+
generator_bn_apply: bool = False,
|
41 |
+
generator_bn_init_weight: float = 30.0,
|
42 |
+
|
43 |
+
phones: list = [],
|
44 |
+
dict_fpath: str = "",
|
45 |
+
special_token_nums: int = 4, # [<s>, <pad>, </s>, <unk>]
|
46 |
+
**kwargs
|
47 |
+
):
|
48 |
+
super().__init__(**kwargs)
|
49 |
+
# read in all the configurations
|
50 |
+
self.segmenter_type = segmenter_type
|
51 |
+
self.segmenter_input_dim = segmenter_input_dim
|
52 |
+
self.segmenter_hidden_dim = segmenter_hidden_dim
|
53 |
+
self.segmenter_dropout = segmenter_dropout
|
54 |
+
self.segmenter_kernel_size = segmenter_kernel_size
|
55 |
+
|
56 |
+
self.discriminator_input_dim = discriminator_input_dim
|
57 |
+
self.discriminator_kernel = discriminator_kernel
|
58 |
+
self.discriminator_dilation = discriminator_dilation
|
59 |
+
self.discriminator_dim = discriminator_dim
|
60 |
+
self.discriminator_causal = discriminator_causal
|
61 |
+
self.discriminator_linear_emb = discriminator_linear_emb
|
62 |
+
self.discriminator_depth = discriminator_depth
|
63 |
+
self.discriminator_max_pool = discriminator_max_pool
|
64 |
+
self.discriminator_act_after_linear = discriminator_act_after_linear
|
65 |
+
self.discriminator_dropout = discriminator_dropout
|
66 |
+
self.discriminator_spectral_norm = discriminator_spectral_norm
|
67 |
+
self.discriminator_weight_norm = discriminator_weight_norm
|
68 |
+
|
69 |
+
self.generator_input_dim = generator_input_dim
|
70 |
+
self.generator_output_dim = generator_output_dim
|
71 |
+
self.generator_kernel = generator_kernel
|
72 |
+
self.generator_dilation = generator_dilation
|
73 |
+
self.generator_stride = generator_stride
|
74 |
+
self.generator_bias = generator_bias
|
75 |
+
self.generator_dropout = generator_dropout
|
76 |
+
self.generator_bn_apply = generator_bn_apply
|
77 |
+
self.generator_bn_init_weight = generator_bn_init_weight
|
78 |
+
|
79 |
+
self.special_token_nums = special_token_nums
|
80 |
+
if os.path.isfile(dict_fpath):
|
81 |
+
self.phones = self.read_phns_dict_from_fpath(dict_fpath)
|
82 |
+
else:
|
83 |
+
self.phones = phones
|
84 |
+
if len(self.phones) > 0:
|
85 |
+
self.generator_output_dim = len(self.phones) + self.special_token_nums
|
86 |
+
self.discriminator_input_dim = self.generator_output_dim
|
87 |
+
|
88 |
+
def read_phns_dict_from_fpath(self, fpath: str):
|
89 |
+
phns = []
|
90 |
+
with open(fpath, "r", encoding="utf-8") as f:
|
91 |
+
for l in f:
|
92 |
+
phn = l.strip().split('\t')[0].split(' ')[0]
|
93 |
+
phns.append(phn)
|
94 |
+
return phns
|
95 |
+
|
96 |
+
def main():
|
97 |
+
config = RebornUASRConfig(dict_fpath="/home/andybi7676/Desktop/uasr-rl/data2/nl_mls/text/prep/phones/dict.phn.txt")
|
98 |
+
print(config)
|
99 |
+
output_fpath = "./reborn_uasr_configs/config_mls-nl.json"
|
100 |
+
with open(output_fpath, 'w', encoding='utf-8') as fw:
|
101 |
+
config_json_string = json.dumps(config.to_dict(), indent=2, sort_keys=True) + "\n"
|
102 |
+
fw.write(config_json_string)
|
103 |
+
|
104 |
+
if __name__ == "__main__":
|
105 |
+
main()
|
modeling_reborn.py
ADDED
@@ -0,0 +1,381 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
import torch.nn as nn
|
3 |
+
from transformers import PreTrainedModel
|
4 |
+
from .configuration_reborn import RebornUASRConfig
|
5 |
+
from typing import Optional, Tuple, Union, List
|
6 |
+
|
7 |
+
class RebornSegmenter(nn.Module):
|
8 |
+
def __init__(self, config):
|
9 |
+
super().__init__()
|
10 |
+
self.config = config
|
11 |
+
self.conv1 = nn.Conv1d(config.segmenter_input_dim, config.segmenter_hidden_dim, config.segmenter_kernel_size, padding=config.segmenter_kernel_size//2)
|
12 |
+
self.conv2 = nn.Conv1d(config.segmenter_hidden_dim, config.segmenter_hidden_dim, 3, padding=1)
|
13 |
+
self.conv3 = nn.Conv1d(config.segmenter_hidden_dim, 2, 1)
|
14 |
+
self.dropout = nn.Dropout(config.segmenter_dropout)
|
15 |
+
self.relu = nn.ReLU()
|
16 |
+
|
17 |
+
def forward(self, x):
|
18 |
+
"""
|
19 |
+
Input:
|
20 |
+
x: (B, T, C)
|
21 |
+
padding_mask: (B, T) # 0: not padding; 1: padding
|
22 |
+
Output:
|
23 |
+
boundary: (B, T, 2) # 0: not boundary; 1: boundary
|
24 |
+
"""
|
25 |
+
x = x.transpose(1, 2)
|
26 |
+
x = self.dropout(self.relu(self.conv1(x)))
|
27 |
+
x = self.dropout(self.relu(self.conv2(x)))
|
28 |
+
x = self.conv3(x)
|
29 |
+
x = x.transpose(1, 2)
|
30 |
+
return x
|
31 |
+
|
32 |
+
def boundary_predict(self, x, padding_mask, deterministic=False):
|
33 |
+
"""
|
34 |
+
Input:
|
35 |
+
x: (B, T, C)
|
36 |
+
padding_mask: (B, T)
|
37 |
+
Output:
|
38 |
+
boundary: (B, T) # 0: not boundary; 1: boundary
|
39 |
+
boundary_logits: (B, T, 2) # 0: not boundary; 1: boundary
|
40 |
+
"""
|
41 |
+
boundary_logits = self.forward(x)
|
42 |
+
if deterministic:
|
43 |
+
boundary = boundary_logits.argmax(-1)
|
44 |
+
boundary[padding_mask] = -1
|
45 |
+
else:
|
46 |
+
boundary = torch.distributions.Categorical(logits=boundary_logits).sample()
|
47 |
+
boundary[padding_mask] = -1
|
48 |
+
return boundary, boundary_logits
|
49 |
+
|
50 |
+
def pre_segment(self, logits, padding_mask, return_boundary=False, deterministic=True):
|
51 |
+
"""
|
52 |
+
Input:
|
53 |
+
logits: (B, T, C)
|
54 |
+
padding_mask: (B, T)
|
55 |
+
Output:
|
56 |
+
new_logits: (B, T', C)
|
57 |
+
new_padding_mask: (B, T')
|
58 |
+
"""
|
59 |
+
|
60 |
+
bsz, tsz, csz = logits.size()
|
61 |
+
|
62 |
+
boundary, boundary_logits = self.boundary_predict(logits, padding_mask, deterministic=deterministic)
|
63 |
+
|
64 |
+
# max boundary number
|
65 |
+
# print("boundary", boundary)
|
66 |
+
# print(torch.sum(boundary==1, dim=1))
|
67 |
+
new_tsz = int(torch.max(torch.sum(boundary==1, dim=1)).item())+1 # add <bos>
|
68 |
+
new_logits = logits.new_zeros(bsz, new_tsz, csz)
|
69 |
+
new_pad = padding_mask.new_zeros(bsz, new_tsz)
|
70 |
+
|
71 |
+
for b in range(bsz):
|
72 |
+
# merge consecutive segments when meeting a boundary (mean_pool_join)
|
73 |
+
new_idx = 0
|
74 |
+
count = 0
|
75 |
+
for t in range(tsz):
|
76 |
+
if padding_mask[b, t] == 1:
|
77 |
+
break
|
78 |
+
if boundary[b, t] == 1:
|
79 |
+
new_logits[b, new_idx] /= count
|
80 |
+
new_idx += 1
|
81 |
+
count = 0
|
82 |
+
new_logits[b, new_idx] += logits[b, t]
|
83 |
+
count += 1
|
84 |
+
if count > 0:
|
85 |
+
# last segment
|
86 |
+
new_logits[b, new_idx] /= count
|
87 |
+
new_idx += 1
|
88 |
+
count = 0
|
89 |
+
if new_idx < new_tsz:
|
90 |
+
pad = new_tsz - new_idx
|
91 |
+
new_logits[b, -pad:] = 0
|
92 |
+
new_pad[b, -pad:] = True
|
93 |
+
|
94 |
+
if return_boundary:
|
95 |
+
return new_logits, new_pad, boundary, boundary_logits
|
96 |
+
return new_logits, new_pad
|
97 |
+
|
98 |
+
class RebornGenerator(nn.Module):
|
99 |
+
def __init__(self, config):
|
100 |
+
super().__init__()
|
101 |
+
|
102 |
+
self.config = config
|
103 |
+
self.output_dim = config.generator_output_dim
|
104 |
+
self.stride = config.generator_stride
|
105 |
+
self.dropout = nn.Dropout(config.generator_dropout)
|
106 |
+
cnn_input_dim = config.generator_input_dim
|
107 |
+
cnn_output_dim = config.generator_output_dim
|
108 |
+
|
109 |
+
padding = config.generator_kernel // 2
|
110 |
+
self.proj = nn.Sequential(
|
111 |
+
nn.Conv1d(
|
112 |
+
cnn_input_dim,
|
113 |
+
cnn_output_dim,
|
114 |
+
kernel_size=config.generator_kernel,
|
115 |
+
stride=config.generator_stride,
|
116 |
+
dilation=config.generator_dilation,
|
117 |
+
padding=padding,
|
118 |
+
bias=config.generator_bias,
|
119 |
+
),
|
120 |
+
)
|
121 |
+
|
122 |
+
def forward(self, dense_x, tokens, dense_padding_mask):
|
123 |
+
dense_x = self.dropout(dense_x)
|
124 |
+
# (B, T, C) -> (B, C, T)
|
125 |
+
dense_x = dense_x.transpose(-2, -1)
|
126 |
+
|
127 |
+
dense_x = self.proj(dense_x)
|
128 |
+
# (B, C, T) -> (B, T, C)
|
129 |
+
dense_x = dense_x.transpose(-2, -1)
|
130 |
+
if self.stride > 1:
|
131 |
+
dense_padding_mask = dense_padding_mask[:, :: self.stride]
|
132 |
+
|
133 |
+
if dense_padding_mask.size(1) != dense_x.size(1):
|
134 |
+
new_padding = dense_padding_mask.new_zeros(dense_x.shape[:-1])
|
135 |
+
diff = new_padding.size(1) - dense_padding_mask.size(1)
|
136 |
+
assert (
|
137 |
+
diff > 0
|
138 |
+
), f"{new_padding.shape}, {dense_padding_mask.shape}, {dense_x.shape}, {diff}"
|
139 |
+
if diff > 0:
|
140 |
+
new_padding[:, diff:] = dense_padding_mask
|
141 |
+
else:
|
142 |
+
assert diff < 0
|
143 |
+
new_padding = dense_padding_mask[:, :diff]
|
144 |
+
|
145 |
+
dense_padding_mask = new_padding
|
146 |
+
|
147 |
+
result = {}
|
148 |
+
|
149 |
+
token_x = None
|
150 |
+
if tokens is not None:
|
151 |
+
token_x = dense_x.new_zeros(tokens.numel(), self.output_dim)
|
152 |
+
token_x.scatter_(1, tokens.view(-1, 1).long(), 1)
|
153 |
+
token_x = token_x.view(tokens.shape + (self.output_dim,))
|
154 |
+
|
155 |
+
result["dense_x"] = dense_x
|
156 |
+
result["token_x"] = token_x
|
157 |
+
result["dense_padding_mask"] = dense_padding_mask
|
158 |
+
|
159 |
+
return result
|
160 |
+
|
161 |
+
def get_item(tensor):
|
162 |
+
# tpu-comment: making this a no-op for xla devices.
|
163 |
+
if torch.is_tensor(tensor) and tensor.device.type == "xla":
|
164 |
+
return tensor.detach()
|
165 |
+
if hasattr(tensor, "item"):
|
166 |
+
return tensor.item()
|
167 |
+
if hasattr(tensor, "__getitem__"):
|
168 |
+
return tensor[0]
|
169 |
+
return tensor
|
170 |
+
|
171 |
+
def post_process(sentence: str, symbol: str):
|
172 |
+
if symbol == "sentencepiece":
|
173 |
+
sentence = sentence.replace(" ", "").replace("\u2581", " ").strip()
|
174 |
+
elif symbol == "wordpiece":
|
175 |
+
sentence = sentence.replace(" ", "").replace("_", " ").strip()
|
176 |
+
elif symbol == "letter":
|
177 |
+
sentence = sentence.replace(" ", "").replace("|", " ").strip()
|
178 |
+
elif symbol == "silence":
|
179 |
+
import re
|
180 |
+
sentence = sentence.replace("<SIL>", "")
|
181 |
+
sentence = re.sub(' +', ' ', sentence).strip()
|
182 |
+
elif symbol == "_EOW":
|
183 |
+
sentence = sentence.replace(" ", "").replace("_EOW", " ").strip()
|
184 |
+
elif symbol in {"subword_nmt", "@@ ", "@@"}:
|
185 |
+
if symbol == "subword_nmt":
|
186 |
+
symbol = "@@ "
|
187 |
+
sentence = (sentence + " ").replace(symbol, "").rstrip()
|
188 |
+
elif symbol == "none":
|
189 |
+
pass
|
190 |
+
elif symbol is not None:
|
191 |
+
raise NotImplementedError(f"Unknown post_process option: {symbol}")
|
192 |
+
return sentence
|
193 |
+
|
194 |
+
class SimpleTokenizer(object):
|
195 |
+
def __init__(self,
|
196 |
+
phones: List[str],
|
197 |
+
bos="<s>",
|
198 |
+
pad="<pad>",
|
199 |
+
eos="</s>",
|
200 |
+
unk="<unk>",
|
201 |
+
extra_special_symbols=None,
|
202 |
+
) -> None:
|
203 |
+
self.bos_word, self.unk_word, self.pad_word, self.eos_word = bos, unk, pad, eos
|
204 |
+
self.symbols = []
|
205 |
+
self.count = []
|
206 |
+
self.indices = {}
|
207 |
+
self.bos_index = self.add_symbol(bos)
|
208 |
+
self.pad_index = self.add_symbol(pad)
|
209 |
+
self.eos_index = self.add_symbol(eos)
|
210 |
+
self.unk_index = self.add_symbol(unk)
|
211 |
+
if extra_special_symbols:
|
212 |
+
for s in extra_special_symbols:
|
213 |
+
self.add_symbol(s)
|
214 |
+
self.nspecial = len(self.symbols)
|
215 |
+
for phone in phones:
|
216 |
+
self.add_symbol(phone)
|
217 |
+
self.postprocess_code = "silence"
|
218 |
+
|
219 |
+
def add_symbol(self, word, n=1, overwrite=False):
|
220 |
+
"""Adds a word to the dictionary"""
|
221 |
+
if word in self.indices and not overwrite:
|
222 |
+
idx = self.indices[word]
|
223 |
+
self.count[idx] = self.count[idx] + n
|
224 |
+
return idx
|
225 |
+
else:
|
226 |
+
idx = len(self.symbols)
|
227 |
+
self.indices[word] = idx
|
228 |
+
self.symbols.append(word)
|
229 |
+
self.count.append(n)
|
230 |
+
return idx
|
231 |
+
|
232 |
+
def __eq__(self, other):
|
233 |
+
return self.indices == other.indices
|
234 |
+
|
235 |
+
def __getitem__(self, idx):
|
236 |
+
if idx < len(self.symbols):
|
237 |
+
return self.symbols[idx]
|
238 |
+
return self.unk_word
|
239 |
+
|
240 |
+
def get_count(self, idx):
|
241 |
+
return self.count[idx]
|
242 |
+
|
243 |
+
def __len__(self):
|
244 |
+
"""Returns the number of symbols in the dictionary"""
|
245 |
+
return len(self.symbols)
|
246 |
+
|
247 |
+
def __contains__(self, sym):
|
248 |
+
return sym in self.indices
|
249 |
+
|
250 |
+
def index(self, sym):
|
251 |
+
"""Returns the index of the specified symbol"""
|
252 |
+
assert isinstance(sym, str)
|
253 |
+
if sym in self.indices:
|
254 |
+
return self.indices[sym]
|
255 |
+
return self.unk_index
|
256 |
+
|
257 |
+
def string(
|
258 |
+
self,
|
259 |
+
tensor,
|
260 |
+
bpe_symbol=None,
|
261 |
+
escape_unk=False,
|
262 |
+
extra_symbols_to_ignore=None,
|
263 |
+
unk_string=None,
|
264 |
+
include_eos=False,
|
265 |
+
separator=" ",
|
266 |
+
):
|
267 |
+
"""Helper for converting a tensor of token indices to a string.
|
268 |
+
|
269 |
+
Can optionally remove BPE symbols or escape <unk> words.
|
270 |
+
"""
|
271 |
+
if torch.is_tensor(tensor) and tensor.dim() == 2:
|
272 |
+
return "\n".join(
|
273 |
+
self.string(
|
274 |
+
t,
|
275 |
+
bpe_symbol,
|
276 |
+
escape_unk,
|
277 |
+
extra_symbols_to_ignore,
|
278 |
+
include_eos=include_eos,
|
279 |
+
)
|
280 |
+
for t in tensor
|
281 |
+
)
|
282 |
+
|
283 |
+
extra_symbols_to_ignore = set(extra_symbols_to_ignore or [])
|
284 |
+
if not include_eos:
|
285 |
+
extra_symbols_to_ignore.add(self.eos())
|
286 |
+
|
287 |
+
def token_string(i):
|
288 |
+
if i == self.unk():
|
289 |
+
if unk_string is not None:
|
290 |
+
return unk_string
|
291 |
+
else:
|
292 |
+
return self.unk_string(escape_unk)
|
293 |
+
else:
|
294 |
+
return self[i]
|
295 |
+
|
296 |
+
if hasattr(self, "bos_index"):
|
297 |
+
extra_symbols_to_ignore.add(self.bos())
|
298 |
+
|
299 |
+
sent = separator.join(
|
300 |
+
token_string(i)
|
301 |
+
for i in tensor
|
302 |
+
if get_item(i) not in extra_symbols_to_ignore
|
303 |
+
)
|
304 |
+
|
305 |
+
return post_process(sent, bpe_symbol)
|
306 |
+
|
307 |
+
def unk_string(self, escape=False):
|
308 |
+
"""Return unknown string, optionally escaped as: <<unk>>"""
|
309 |
+
if escape:
|
310 |
+
return "<{}>".format(self.unk_word)
|
311 |
+
else:
|
312 |
+
return self.unk_word
|
313 |
+
|
314 |
+
def bos(self):
|
315 |
+
"""Helper to get index of beginning-of-sentence symbol"""
|
316 |
+
return self.bos_index
|
317 |
+
|
318 |
+
def pad(self):
|
319 |
+
"""Helper to get index of pad symbol"""
|
320 |
+
return self.pad_index
|
321 |
+
|
322 |
+
def eos(self):
|
323 |
+
"""Helper to get index of end-of-sentence symbol"""
|
324 |
+
return self.eos_index
|
325 |
+
|
326 |
+
def unk(self):
|
327 |
+
"""Helper to get index of unk symbol"""
|
328 |
+
return self.unk_index
|
329 |
+
|
330 |
+
|
331 |
+
class RebornUASRModel(PreTrainedModel):
|
332 |
+
config_class = RebornUASRConfig
|
333 |
+
|
334 |
+
def __init__(self, config):
|
335 |
+
super().__init__(config)
|
336 |
+
self.pca = nn.Linear(1024, 512)
|
337 |
+
self.segmenter = RebornSegmenter(config)
|
338 |
+
self.generator = RebornGenerator(config)
|
339 |
+
self.tokenizer = None
|
340 |
+
if len(config.phones) > 0:
|
341 |
+
self.tokenizer = SimpleTokenizer(config.phones)
|
342 |
+
|
343 |
+
def forward(
|
344 |
+
self,
|
345 |
+
x: Optional[torch.Tensor], # (B, T, C)
|
346 |
+
padding_mask: Optional[torch.Tensor], # (B, T)
|
347 |
+
):
|
348 |
+
x_reduced = self.pca(x)
|
349 |
+
x_segmented, segmented_padding_mask = self.segmenter.pre_segment(x_reduced, padding_mask, deterministic=True)
|
350 |
+
x_generated = self.generator(x_segmented, None, segmented_padding_mask)
|
351 |
+
|
352 |
+
return {
|
353 |
+
'x_reduced': x_reduced,
|
354 |
+
'x_segmented': x_segmented,
|
355 |
+
'x_generated': x_generated
|
356 |
+
}
|
357 |
+
|
358 |
+
def generate(self, x, padding_mask, merge_consecutive=True, remove_silence=True):
|
359 |
+
res = self.forward(x, padding_mask)
|
360 |
+
y_raw_logits = res['x_generated']['dense_x']
|
361 |
+
y_raw_padding = res['x_generated']['dense_padding_mask']
|
362 |
+
y_raw_logits[y_raw_padding][..., self.tokenizer.pad_index] = float('inf')
|
363 |
+
preds = y_raw_logits.argmax(-1)
|
364 |
+
hyps = []
|
365 |
+
postprocess_code = "silence" if remove_silence else "none"
|
366 |
+
for pred in preds:
|
367 |
+
if merge_consecutive:
|
368 |
+
# merge consecutive predictions
|
369 |
+
pred = torch.unique_consecutive(pred)
|
370 |
+
hyp = self.tokenizer.string(pred, bpe_symbol=postprocess_code)
|
371 |
+
hyps.append(hyp)
|
372 |
+
return hyps
|
373 |
+
|
374 |
+
def main():
|
375 |
+
model_config = RebornUASRConfig.from_pretrained("/home/andybi7676/Desktop/uasr-rl/reborn_uasr/config.json")
|
376 |
+
print(model_config)
|
377 |
+
model = RebornUASRModel(model_config)
|
378 |
+
print(model.tokenizer.indices)
|
379 |
+
|
380 |
+
if __name__ == "__main__":
|
381 |
+
main()
|
pytorch_model.bin
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:fee32077a74c336da28274df0fc68e212952baa860e0cc4a10125430ce096eab
|
3 |
+
size 12973069
|