cantabile-kwok
commited on
Upload config.yml
Browse files- config.yml +201 -0
config.yml
ADDED
@@ -0,0 +1,201 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
allow_cache: false
|
2 |
+
batch_frames: 3600
|
3 |
+
config: conf/ctxv2w.v1.yaml
|
4 |
+
crop_max_frames: 100
|
5 |
+
discriminator_adv_loss_params:
|
6 |
+
average_by_discriminators: false
|
7 |
+
discriminator_grad_norm: -1
|
8 |
+
discriminator_optimizer_params:
|
9 |
+
betas:
|
10 |
+
- 0.5
|
11 |
+
- 0.9
|
12 |
+
lr: 0.0002
|
13 |
+
weight_decay: 0.0
|
14 |
+
discriminator_optimizer_type: Adam
|
15 |
+
discriminator_params:
|
16 |
+
follow_official_norm: true
|
17 |
+
period_discriminator_params:
|
18 |
+
bias: true
|
19 |
+
channels: 32
|
20 |
+
downsample_scales:
|
21 |
+
- 3
|
22 |
+
- 3
|
23 |
+
- 3
|
24 |
+
- 3
|
25 |
+
- 1
|
26 |
+
in_channels: 1
|
27 |
+
kernel_sizes:
|
28 |
+
- 5
|
29 |
+
- 3
|
30 |
+
max_downsample_channels: 1024
|
31 |
+
nonlinear_activation: LeakyReLU
|
32 |
+
nonlinear_activation_params:
|
33 |
+
negative_slope: 0.1
|
34 |
+
out_channels: 1
|
35 |
+
use_spectral_norm: false
|
36 |
+
use_weight_norm: true
|
37 |
+
periods:
|
38 |
+
- 2
|
39 |
+
- 3
|
40 |
+
- 5
|
41 |
+
- 7
|
42 |
+
- 11
|
43 |
+
scale_discriminator_params:
|
44 |
+
bias: true
|
45 |
+
channels: 128
|
46 |
+
downsample_scales:
|
47 |
+
- 4
|
48 |
+
- 4
|
49 |
+
- 4
|
50 |
+
- 4
|
51 |
+
- 1
|
52 |
+
in_channels: 1
|
53 |
+
kernel_sizes:
|
54 |
+
- 15
|
55 |
+
- 41
|
56 |
+
- 5
|
57 |
+
- 3
|
58 |
+
max_downsample_channels: 1024
|
59 |
+
max_groups: 16
|
60 |
+
nonlinear_activation: LeakyReLU
|
61 |
+
nonlinear_activation_params:
|
62 |
+
negative_slope: 0.1
|
63 |
+
out_channels: 1
|
64 |
+
scale_downsample_pooling: AvgPool1d
|
65 |
+
scale_downsample_pooling_params:
|
66 |
+
kernel_size: 4
|
67 |
+
padding: 2
|
68 |
+
stride: 2
|
69 |
+
scales: 3
|
70 |
+
discriminator_scheduler_params:
|
71 |
+
gamma: 0.5
|
72 |
+
milestones:
|
73 |
+
- 200000
|
74 |
+
- 400000
|
75 |
+
- 600000
|
76 |
+
- 800000
|
77 |
+
discriminator_scheduler_type: MultiStepLR
|
78 |
+
discriminator_train_start_steps: 0
|
79 |
+
discriminator_type: HiFiGANMultiScaleMultiPeriodDiscriminator
|
80 |
+
distributed: true
|
81 |
+
dropout_features: 0.0
|
82 |
+
eval_interval_steps: 100000
|
83 |
+
feat_match_loss_params:
|
84 |
+
average_by_discriminators: false
|
85 |
+
average_by_layers: false
|
86 |
+
include_final_outputs: false
|
87 |
+
frontend_mel_prediction_stop_steps: 200000
|
88 |
+
frontend_params:
|
89 |
+
conformer_params:
|
90 |
+
activation_type: swish
|
91 |
+
attention_dim: 184
|
92 |
+
attention_dropout_rate: 0.2
|
93 |
+
attention_heads: 2
|
94 |
+
cnn_module_kernel: 31
|
95 |
+
concat_after: false
|
96 |
+
dropout_rate: 0.2
|
97 |
+
linear_units: 1536
|
98 |
+
macaron_style: true
|
99 |
+
normalize_before: true
|
100 |
+
num_blocks: 2
|
101 |
+
pos_enc_layer_type: rel_pos
|
102 |
+
positional_dropout_rate: 0.2
|
103 |
+
positionwise_conv_kernel_size: 3
|
104 |
+
positionwise_layer_type: conv1d
|
105 |
+
selfattention_layer_type: rel_selfattn
|
106 |
+
use_cnn_module: true
|
107 |
+
prompt_channels: 1024
|
108 |
+
vqvec_channels: 512
|
109 |
+
generator_adv_loss_params:
|
110 |
+
average_by_discriminators: false
|
111 |
+
generator_grad_norm: -1
|
112 |
+
generator_optimizer_params:
|
113 |
+
betas:
|
114 |
+
- 0.5
|
115 |
+
- 0.9
|
116 |
+
lr: 0.0002
|
117 |
+
weight_decay: 0.0
|
118 |
+
generator_optimizer_type: Adam
|
119 |
+
generator_params:
|
120 |
+
bias: true
|
121 |
+
channels: 512
|
122 |
+
condition_dim: 1024
|
123 |
+
in_channels: 184
|
124 |
+
kernel_size: 7
|
125 |
+
nonlinear_activation: snakebeta-condition
|
126 |
+
out_channels: 1
|
127 |
+
resblock: '1'
|
128 |
+
resblock_dilations:
|
129 |
+
- - 1
|
130 |
+
- 3
|
131 |
+
- 5
|
132 |
+
- - 1
|
133 |
+
- 3
|
134 |
+
- 5
|
135 |
+
- - 1
|
136 |
+
- 3
|
137 |
+
- 5
|
138 |
+
resblock_kernel_sizes:
|
139 |
+
- 3
|
140 |
+
- 7
|
141 |
+
- 11
|
142 |
+
snake_logscale: true
|
143 |
+
upsample_kernel_sizes:
|
144 |
+
- 16
|
145 |
+
- 10
|
146 |
+
- 6
|
147 |
+
- 4
|
148 |
+
upsample_scales:
|
149 |
+
- 8
|
150 |
+
- 5
|
151 |
+
- 3
|
152 |
+
- 2
|
153 |
+
use_additional_convs: true
|
154 |
+
use_weight_norm: true
|
155 |
+
generator_scheduler_params:
|
156 |
+
gamma: 0.5
|
157 |
+
milestones:
|
158 |
+
- 200000
|
159 |
+
- 400000
|
160 |
+
- 600000
|
161 |
+
- 800000
|
162 |
+
generator_scheduler_type: MultiStepLR
|
163 |
+
generator_train_start_steps: 1
|
164 |
+
generator_type: BigVGAN
|
165 |
+
hop_size: 240
|
166 |
+
lambda_adv: 1.0
|
167 |
+
lambda_aux: 45.0
|
168 |
+
lambda_feat_match: 2.0
|
169 |
+
lambda_frontend_mel_prediction: 60
|
170 |
+
log_interval_steps: 1000
|
171 |
+
max_num_frames: 3000
|
172 |
+
mel_loss_params:
|
173 |
+
fft_size: 2048
|
174 |
+
fmax: 8000
|
175 |
+
fmin: 40
|
176 |
+
fs: 24000
|
177 |
+
hop_size: 300
|
178 |
+
log_base: null
|
179 |
+
num_mels: 80
|
180 |
+
win_length: 1200
|
181 |
+
window: hann
|
182 |
+
min_num_frames: 600
|
183 |
+
num_mels: 80
|
184 |
+
num_save_intermediate_results: 4
|
185 |
+
num_workers: 8
|
186 |
+
outdir: exp/train_all_ctxv2w.v1
|
187 |
+
pin_memory: true
|
188 |
+
pretrain: ''
|
189 |
+
prompt_fold_by_2: true
|
190 |
+
prompt_net_type: ConvPromptPrenet
|
191 |
+
rank: 0
|
192 |
+
sampling_rate: 24000
|
193 |
+
save_interval_steps: 10000
|
194 |
+
use_feat_match_loss: true
|
195 |
+
use_mel_loss: true
|
196 |
+
use_stft_loss: false
|
197 |
+
verbose: 1
|
198 |
+
version: 0.5.3
|
199 |
+
vq_codebook: feats/vqidx/codebook.npy
|
200 |
+
win_length: 697
|
201 |
+
world_size: 4
|