Tacotron2/hparams.py at main · BridgetteSong/Tacotron2 · GitHub

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

from text import splittoken2index, alltoken2index
import os

# CONFIG -----------------------------------------------------------------------------------------------------------#

# File Path
wav_path = "wavs/all_16k"
text_dict_path = "dict_text.pkl"


# Setting Saving path
phones_path = "phone_text.txt"
checkpoint_path = "checkpoints"
gta_path = os.path.join(checkpoint_path, "gta")


################################
# Audio Parameters             #
################################
max_wav_value=32768.0
sampling_rate=16000
filter_length=1024
hop_length=256
win_length=1024
mel_fmin=80.0
mel_fmax=7600.0
compression = ["log", "log10"][0]  # must be one of them

################################
# Encoder Network parameters   #
################################
split_tone = False
num_chars = len(splittoken2index) if split_tone else len(alltoken2index)
encoder_kernel_size=5
encoder_n_convolutions=3
encoder_embedding_dim=512
tone_embedding = encoder_embedding_dim//8

#############################################
# Reference Encoder Network Hyperparameters #
#############################################
speaker_encoder_type = ["GST", "VAE", "GMVAE"][0]
expressive_encoder_type = ["GST", "VAE", "GMVAE"][0]

spk_ids = {"us": 0,
           "**": 1}
speaker_classes = len(spk_ids)

emotioned = False
emotion_classes = speaker_classes

cat_lambda = 0.0
cat_incr = 0.01
cat_step = 1000
cat_step_after = 20
cat_max_step = 300000

kl_lambda = 0.00001
kl_incr = 0.000001
kl_step = 1000
kl_step_after = 500
kl_max_step = 300000

# reference_encoder
ref_enc_filters=[32, 32, 64, 64, 128, 128]
ref_enc_size=[3, 3]
ref_enc_strides=[2, 2]
ref_enc_pad=[1, 1]
ref_enc_gru_size=128

# Style Token Layer
token_num=10
num_heads=8

# embedding size
token_embedding_size=256
speaker_embedding_size=64
vae_size=32

################################
# Decoder Network parameters   #
################################
feed_back_last=True
n_mel_channels=80
n_frames_per_step=2
decoder_rnn_dim=512
prenet_dims=[256, 256]
gate_threshold=0.5
max_decoder_steps=1000
p_attention_dropout=0.1
p_decoder_dropout=0.1

#################################
# Attention Network parameters  #
#################################
attention_mode=["GMM", "FAV2"][1]
attention_rnn_dim=512
attention_dim=128

# Location Layer parameters
attention_location_n_filters=32
attention_location_kernel_size=31

# GMM parameters
delta_bias=1.0
sigma_bias=10.0
gmm_kernel=5

################################
# Auxiliary Loss parameters    #
################################
guided_sigma=0.2                # weight for guided attention loss. default 0.4
pos_weight=15.0                 # BCEWithLogitsLoss pos_weight, default 10.0

################################
# Mel-post Network parameters  #
################################
postnet_embedding_dims=[512, 512, 512, 512]
postnet_kernel_sizes=[5, 5, 5, 5]
p_postnet_dropout=0.5

postnet_k=5
postnet_num_highways=4
post_projections=[encoder_embedding_dim, encoder_embedding_dim]

################################
# Training parameters          #
################################
distributed_run=True
dist_backend="nccl"
dist_url="tcp://localhost:43021"
seed=1234
dynamic_loss_scaling=True
batch_size=32
learning_rate=1e-3
weight_decay=1e-6
lr_decay=0.999
training_steps=200_000
epochs=500
grad_clip_thresh=1.0
save_checkpoint_every_n_step=1_000 if distributed_run else 10_000
# ------------------------------------------------------------------------------------------------------------------#