Hugging Face's logo

Fhrozen
/
vctk_hifigan.full_band.v1

Audio-to-Audio
ESPnet
English
audio
vocoder
Model card Files Community
Fhrozen commited on
Commit
5731db4
•
1 Parent(s): e6223dd

add files

Browse files
Files changed (3) hide show
  1. checkpoint-2500000steps.pkl +3 -0
  2. config.yml +191 -0
  3. stats.h5 +3 -0
checkpoint-2500000steps.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e182ccc2924ceb2c37d61429cbea06925526eadb94d0788101b52f38208db893
3
+ size 1004891049
config.yml ADDED
@@ -0,0 +1,191 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ allow_cache: true
2
+ batch_max_steps: 8400
3
+ batch_size: 16
4
+ config: conf/hifigan.full_band.v1.yaml
5
+ dev_dumpdir: dump/dev_all/norm
6
+ dev_feats_scp: null
7
+ dev_segments: null
8
+ dev_wav_scp: null
9
+ discriminator_adv_loss_params:
10
+ average_by_discriminators: false
11
+ discriminator_grad_norm: -1
12
+ discriminator_optimizer_params:
13
+ betas:
14
+ - 0.5
15
+ - 0.9
16
+ lr: 0.0002
17
+ weight_decay: 0.0
18
+ discriminator_optimizer_type: Adam
19
+ discriminator_params:
20
+ follow_official_norm: true
21
+ period_discriminator_params:
22
+ bias: true
23
+ channels: 32
24
+ downsample_scales:
25
+ - 3
26
+ - 3
27
+ - 3
28
+ - 3
29
+ - 1
30
+ in_channels: 1
31
+ kernel_sizes:
32
+ - 5
33
+ - 3
34
+ max_downsample_channels: 1024
35
+ nonlinear_activation: LeakyReLU
36
+ nonlinear_activation_params:
37
+ negative_slope: 0.1
38
+ out_channels: 1
39
+ use_spectral_norm: false
40
+ use_weight_norm: true
41
+ periods:
42
+ - 2
43
+ - 3
44
+ - 5
45
+ - 7
46
+ - 11
47
+ scale_discriminator_params:
48
+ bias: true
49
+ channels: 128
50
+ downsample_scales:
51
+ - 4
52
+ - 4
53
+ - 4
54
+ - 4
55
+ - 1
56
+ in_channels: 1
57
+ kernel_sizes:
58
+ - 15
59
+ - 41
60
+ - 5
61
+ - 3
62
+ max_downsample_channels: 1024
63
+ max_groups: 16
64
+ nonlinear_activation: LeakyReLU
65
+ nonlinear_activation_params:
66
+ negative_slope: 0.1
67
+ out_channels: 1
68
+ scale_downsample_pooling: AvgPool1d
69
+ scale_downsample_pooling_params:
70
+ kernel_size: 4
71
+ padding: 2
72
+ stride: 2
73
+ scales: 3
74
+ discriminator_scheduler_params:
75
+ gamma: 0.5
76
+ milestones:
77
+ - 200000
78
+ - 400000
79
+ - 600000
80
+ - 800000
81
+ discriminator_scheduler_type: MultiStepLR
82
+ discriminator_train_start_steps: 0
83
+ discriminator_type: HiFiGANMultiScaleMultiPeriodDiscriminator
84
+ distributed: false
85
+ eval_interval_steps: 1000
86
+ feat_match_loss_params:
87
+ average_by_discriminators: false
88
+ average_by_layers: false
89
+ include_final_outputs: false
90
+ fft_size: 2048
91
+ fmax: 22050
92
+ fmin: 0
93
+ format: hdf5
94
+ generator_adv_loss_params:
95
+ average_by_discriminators: false
96
+ generator_grad_norm: -1
97
+ generator_optimizer_params:
98
+ betas:
99
+ - 0.5
100
+ - 0.9
101
+ lr: 0.0002
102
+ weight_decay: 0.0
103
+ generator_optimizer_type: Adam
104
+ generator_params:
105
+ bias: true
106
+ channels: 512
107
+ in_channels: 80
108
+ kernel_size: 7
109
+ nonlinear_activation: LeakyReLU
110
+ nonlinear_activation_params:
111
+ negative_slope: 0.1
112
+ out_channels: 1
113
+ resblock_dilations:
114
+ - - 1
115
+ - 3
116
+ - 5
117
+ - - 1
118
+ - 3
119
+ - 5
120
+ - - 1
121
+ - 3
122
+ - 5
123
+ resblock_kernel_sizes:
124
+ - 3
125
+ - 7
126
+ - 11
127
+ upsample_kernel_sizes:
128
+ - 10
129
+ - 10
130
+ - 8
131
+ - 6
132
+ upsample_scales:
133
+ - 5
134
+ - 5
135
+ - 4
136
+ - 3
137
+ use_additional_convs: true
138
+ use_weight_norm: true
139
+ generator_scheduler_params:
140
+ gamma: 0.5
141
+ milestones:
142
+ - 200000
143
+ - 400000
144
+ - 600000
145
+ - 800000
146
+ generator_scheduler_type: MultiStepLR
147
+ generator_train_start_steps: 1
148
+ generator_type: HiFiGANGenerator
149
+ global_gain_scale: 1.0
150
+ hop_size: 300
151
+ lambda_adv: 1.0
152
+ lambda_aux: 45.0
153
+ lambda_feat_match: 2.0
154
+ log_interval_steps: 100
155
+ mel_loss_params:
156
+ fft_size: 2048
157
+ fmax: 22050
158
+ fmin: 0
159
+ fs: 44100
160
+ hop_size: 300
161
+ log_base: null
162
+ num_mels: 80
163
+ win_length: 1200
164
+ window: hann
165
+ num_mels: 80
166
+ num_save_intermediate_results: 4
167
+ num_workers: 6
168
+ outdir: exp/train_nodev_all_vctk_hifigan.full_band.v1
169
+ pin_memory: true
170
+ pretrain: ''
171
+ rank: 0
172
+ remove_short_samples: false
173
+ resume: ''
174
+ sampling_rate: 44100
175
+ save_interval_steps: 10000
176
+ train_dumpdir: dump/train_nodev_all/norm
177
+ train_feats_scp: null
178
+ train_max_steps: 2500000
179
+ train_segments: null
180
+ train_wav_scp: null
181
+ trim_frame_size: 1024
182
+ trim_hop_size: 256
183
+ trim_silence: false
184
+ trim_threshold_in_db: 20
185
+ use_feat_match_loss: true
186
+ use_mel_loss: true
187
+ use_stft_loss: false
188
+ verbose: 1
189
+ version: 0.5.5
190
+ win_length: 1200
191
+ window: hann
stats.h5 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:14330e074f8bb64b189726ea69a9c18c550df0398a99f7ee51d090b5d8d823cd
3
+ size 4736