Hugging Face's logo

Spaces:
sarulab-speech
/
UTMOSv2
Running on Zero

App Files Community
1
UTMOSv2
File size: 2,929 Bytes 
b55d767
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
 
from pathlib import Path
from types import SimpleNamespace

from torchvision import transforms

from utmosv2.transform.xymasking import XYMasking

batch_size = 10
num_folds = 5

sr = 16000

preprocess = SimpleNamespace(
    top_db=30, min_seconds=None, save_path=Path("preprocessed_data")
)

split = SimpleNamespace(
    type="sgkf_kind",
    target="mos",
    group="sys_id",
    kind="dataset",
)

external_data = "all"
use_bvcc = True


validation_dataset = "each"

dataset = SimpleNamespace(
    name="multi_spec_ext",
    specs=[
        SimpleNamespace(
            mode="melspec",
            n_fft=4096,
            hop_length=32,
            win_length=4096,
            n_mels=512,
            shape=(512, 512),
            norm=80,
        ),
        SimpleNamespace(
            mode="melspec",
            n_fft=4096,
            hop_length=32,
            win_length=2048,
            n_mels=512,
            shape=(512, 512),
            norm=80,
        ),
        SimpleNamespace(
            mode="melspec",
            n_fft=4096,
            hop_length=32,
            win_length=1024,
            n_mels=512,
            shape=(512, 512),
            norm=80,
        ),
        SimpleNamespace(
            mode="melspec",
            n_fft=4096,
            hop_length=32,
            win_length=512,
            n_mels=512,
            shape=(512, 512),
            norm=80,
        ),
    ],
    spec_frames=SimpleNamespace(
        num_frames=2, frame_sec=1.4, mixup_inner=True, mixup_alpha=0.4, extend="tile"
    ),
)
transform = dict(
    train=transforms.Compose(
        [
            transforms.Resize((512, 512)),
            XYMasking(
                num_masks_x=(0, 2),
                num_masks_y=(0, 2),
                mask_x_length=(10, 40),
                mask_y_length=(10, 30),
                fill_value=0,
                p=0.5,
            ),
            # transforms.ToTensor(),
        ]
    ),
    valid=transforms.Compose(
        [
            transforms.Resize((512, 512)),
            # transforms.ToTensor()
        ]
    ),
)

loss = [
    (SimpleNamespace(name="pairwize_diff", margin=0.2, norm="l1"), 0.7),
    (SimpleNamespace(name="mse"), 0.2),
]

optimizer = SimpleNamespace(name="adamw", lr=1e-3, weight_decay=1e-4)

scheduler = SimpleNamespace(name="cosine", T_max=None, eta_min=1e-7)

model = SimpleNamespace(
    name="multi_spec_ext",
    multi_spec=SimpleNamespace(
        backbone="tf_efficientnetv2_s.in21k_ft_in1k",
        pretrained=True,
        num_classes=1,
        pool_type="catavgmax",
        # feature_height=16,
        atten=True,
        # classifier=None,
    ),
)

run = SimpleNamespace(
    mixup=True,
    mixup_alpha=0.4,
    num_epochs=20,
)

main_metric = "sys_srcc"
id_name = None


inference = SimpleNamespace(
    save_path=Path("preds"),
    submit_save_path=Path("submissions"),
    num_tta=5,
    batch_size=8,
    extend="tile",
)