Upload 14 files
Browse files- Ecapa_tdnn512_USphereface2_Inter_intra/config.yaml +93 -0
- Ecapa_tdnn512_USphereface2_Inter_intra/models/avg_model.pt +3 -0
- Ecapa_tdnn_512_UAAMSoftmax_Inter_intra/.DS_Store +0 -0
- Ecapa_tdnn_512_UAAMSoftmax_Inter_intra/config.yaml +87 -0
- Ecapa_tdnn_512_UAAMSoftmax_Inter_intra/models/avg_model.pt +3 -0
- Ecapa_tdnn_512_UAMSoftmax_Inter_intra/.DS_Store +0 -0
- Ecapa_tdnn_512_UAMSoftmax_Inter_intra/config.yaml +87 -0
- Ecapa_tdnn_512_UAMSoftmax_Inter_intra/models/avg_model.pt +3 -0
- RedimNetB2_UAAMSoftmax_Inter_intra/config.yaml +86 -0
- RedimNetB2_UAAMSoftmax_Inter_intra/models/avg_model.pt +3 -0
- RedimNetB2_USphereface2_Inter_intra/config.yaml +86 -0
- RedimNetB2_USphereface2_Inter_intra/models/avg_model.pt +3 -0
- ResNet34_UAAMSoftmax_Inter_intra/config.yaml +86 -0
- ResNet34_UAAMSoftmax_Inter_intra/models/avg_model.pt +3 -0
Ecapa_tdnn512_USphereface2_Inter_intra/config.yaml
ADDED
|
@@ -0,0 +1,93 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
alpha_schedule:
|
| 2 |
+
alpha_final: 0.3
|
| 3 |
+
alpha_warmup_end: 140
|
| 4 |
+
alpha_warmup_start: 60
|
| 5 |
+
data_type: shard
|
| 6 |
+
dataloader_args:
|
| 7 |
+
batch_size: 512
|
| 8 |
+
drop_last: true
|
| 9 |
+
num_workers: 12
|
| 10 |
+
pin_memory: false
|
| 11 |
+
prefetch_factor: 8
|
| 12 |
+
dataset_args:
|
| 13 |
+
aug_prob: 0.6
|
| 14 |
+
fbank_args:
|
| 15 |
+
dither: 1.0
|
| 16 |
+
frame_length: 25
|
| 17 |
+
frame_shift: 10
|
| 18 |
+
num_mel_bins: 80
|
| 19 |
+
filter: true
|
| 20 |
+
filter_args:
|
| 21 |
+
max_num_frames: 800
|
| 22 |
+
min_num_frames: 100
|
| 23 |
+
frontend: fbank
|
| 24 |
+
num_frms: 200
|
| 25 |
+
resample_rate: 16000
|
| 26 |
+
sample_num_per_epoch: 0
|
| 27 |
+
shuffle: true
|
| 28 |
+
shuffle_args:
|
| 29 |
+
shuffle_size: 2500
|
| 30 |
+
spec_aug: false
|
| 31 |
+
spec_aug_args:
|
| 32 |
+
max_f: 8
|
| 33 |
+
max_t: 10
|
| 34 |
+
num_f_mask: 1
|
| 35 |
+
num_t_mask: 1
|
| 36 |
+
prob: 0.6
|
| 37 |
+
speed_perturb: true
|
| 38 |
+
enable_amp: false
|
| 39 |
+
exp_dir: exp/u_cube_plus/ecapa_u_cube_SphereFace2_ArcGuide_v3
|
| 40 |
+
gpus:
|
| 41 |
+
- 2
|
| 42 |
+
- 3
|
| 43 |
+
log_batch_interval: 100
|
| 44 |
+
loss: CrossEntropyLoss
|
| 45 |
+
loss_args: {}
|
| 46 |
+
margin_scheduler: MarginScheduler
|
| 47 |
+
margin_update:
|
| 48 |
+
epoch_iter: 1066
|
| 49 |
+
final_margin: 0.2
|
| 50 |
+
fix_start_epoch: 40
|
| 51 |
+
increase_start_epoch: 20
|
| 52 |
+
increase_type: exp
|
| 53 |
+
initial_margin: 0.0
|
| 54 |
+
update_margin: true
|
| 55 |
+
model: ECAPA_TDNN_GLOB_c512
|
| 56 |
+
model_args:
|
| 57 |
+
embed_dim: 192
|
| 58 |
+
feat_dim: 80
|
| 59 |
+
pooling_func: U_Cube_XI
|
| 60 |
+
model_init: null
|
| 61 |
+
noise_data: data/musan/lmdb
|
| 62 |
+
num_avg: 10
|
| 63 |
+
num_epochs: 150
|
| 64 |
+
optimizer: SGD
|
| 65 |
+
optimizer_args:
|
| 66 |
+
lr: 0.1
|
| 67 |
+
momentum: 0.9
|
| 68 |
+
nesterov: true
|
| 69 |
+
weight_decay: 0.0001
|
| 70 |
+
projection_args:
|
| 71 |
+
do_lm: false
|
| 72 |
+
easy_margin: false
|
| 73 |
+
embed_dim: 192
|
| 74 |
+
lanbuda: 0.7
|
| 75 |
+
margin_type: C
|
| 76 |
+
num_class: 17982
|
| 77 |
+
project_type: sphereface2_uncertainty_arcguide
|
| 78 |
+
scale: 32
|
| 79 |
+
t: 3
|
| 80 |
+
reverb_data: data/rirs/lmdb
|
| 81 |
+
save_epoch_interval: 10
|
| 82 |
+
scheduler: ExponentialDecrease
|
| 83 |
+
scheduler_args:
|
| 84 |
+
epoch_iter: 1066
|
| 85 |
+
final_lr: 5.0e-05
|
| 86 |
+
initial_lr: 0.1
|
| 87 |
+
num_epochs: 150
|
| 88 |
+
scale_ratio: 16.0
|
| 89 |
+
warm_from_zero: true
|
| 90 |
+
warm_up_epoch: 6
|
| 91 |
+
seed: 42
|
| 92 |
+
train_data: data/vox2_dev/shard.list
|
| 93 |
+
train_label: data/vox2_dev/utt2spk
|
Ecapa_tdnn512_USphereface2_Inter_intra/models/avg_model.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:ffd4697ddc12b28e325b311e1417bc27c8fbb47934e14e2905eb4854f4053bca
|
| 3 |
+
size 40689419
|
Ecapa_tdnn_512_UAAMSoftmax_Inter_intra/.DS_Store
ADDED
|
Binary file (6.15 kB). View file
|
|
|
Ecapa_tdnn_512_UAAMSoftmax_Inter_intra/config.yaml
ADDED
|
@@ -0,0 +1,87 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
checkpoint: exp/u_cube_plus/ecapa_u_cube_inter_intra/models/model_140.pt
|
| 2 |
+
data_type: shard
|
| 3 |
+
dataloader_args:
|
| 4 |
+
batch_size: 512
|
| 5 |
+
drop_last: true
|
| 6 |
+
num_workers: 12
|
| 7 |
+
pin_memory: false
|
| 8 |
+
prefetch_factor: 8
|
| 9 |
+
dataset_args:
|
| 10 |
+
aug_prob: 0.6
|
| 11 |
+
fbank_args:
|
| 12 |
+
dither: 1.0
|
| 13 |
+
frame_length: 25
|
| 14 |
+
frame_shift: 10
|
| 15 |
+
num_mel_bins: 80
|
| 16 |
+
filter: true
|
| 17 |
+
filter_args:
|
| 18 |
+
max_num_frames: 800
|
| 19 |
+
min_num_frames: 100
|
| 20 |
+
frontend: fbank
|
| 21 |
+
num_frms: 200
|
| 22 |
+
resample_rate: 16000
|
| 23 |
+
sample_num_per_epoch: 0
|
| 24 |
+
shuffle: true
|
| 25 |
+
shuffle_args:
|
| 26 |
+
shuffle_size: 2500
|
| 27 |
+
spec_aug: false
|
| 28 |
+
spec_aug_args:
|
| 29 |
+
max_f: 8
|
| 30 |
+
max_t: 10
|
| 31 |
+
num_f_mask: 1
|
| 32 |
+
num_t_mask: 1
|
| 33 |
+
prob: 0.6
|
| 34 |
+
speed_perturb: true
|
| 35 |
+
enable_amp: false
|
| 36 |
+
exp_dir: exp/u_cube_plus/ecapa_u_cube_inter_intra
|
| 37 |
+
gpus:
|
| 38 |
+
- 2
|
| 39 |
+
- 3
|
| 40 |
+
log_batch_interval: 100
|
| 41 |
+
loss: CrossEntropyLoss
|
| 42 |
+
loss_args: {}
|
| 43 |
+
margin_scheduler: MarginScheduler
|
| 44 |
+
margin_update:
|
| 45 |
+
epoch_iter: 1066
|
| 46 |
+
final_margin: 0.2
|
| 47 |
+
fix_start_epoch: 40
|
| 48 |
+
increase_start_epoch: 20
|
| 49 |
+
increase_type: exp
|
| 50 |
+
initial_margin: 0.0
|
| 51 |
+
update_margin: true
|
| 52 |
+
model: ECAPA_TDNN_GLOB_c512
|
| 53 |
+
model_args:
|
| 54 |
+
embed_dim: 192
|
| 55 |
+
feat_dim: 80
|
| 56 |
+
pooling_func: U_Cube_XI
|
| 57 |
+
model_init: null
|
| 58 |
+
noise_data: data/musan/lmdb
|
| 59 |
+
num_avg: 10
|
| 60 |
+
num_epochs: 150
|
| 61 |
+
optimizer: SGD
|
| 62 |
+
optimizer_args:
|
| 63 |
+
lr: 0.1
|
| 64 |
+
momentum: 0.9
|
| 65 |
+
nesterov: true
|
| 66 |
+
weight_decay: 0.0001
|
| 67 |
+
projection_args:
|
| 68 |
+
do_lm: false
|
| 69 |
+
easy_margin: false
|
| 70 |
+
embed_dim: 192
|
| 71 |
+
num_class: 17982
|
| 72 |
+
project_type: arc_margin_uncertainty_inter_intra
|
| 73 |
+
scale: 32
|
| 74 |
+
reverb_data: data/rirs/lmdb
|
| 75 |
+
save_epoch_interval: 10
|
| 76 |
+
scheduler: ExponentialDecrease
|
| 77 |
+
scheduler_args:
|
| 78 |
+
epoch_iter: 1066
|
| 79 |
+
final_lr: 5.0e-05
|
| 80 |
+
initial_lr: 0.1
|
| 81 |
+
num_epochs: 150
|
| 82 |
+
scale_ratio: 16.0
|
| 83 |
+
warm_from_zero: true
|
| 84 |
+
warm_up_epoch: 6
|
| 85 |
+
seed: 42
|
| 86 |
+
train_data: data/vox2_dev/shard.list
|
| 87 |
+
train_label: data/vox2_dev/utt2spk
|
Ecapa_tdnn_512_UAAMSoftmax_Inter_intra/models/avg_model.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:922d93a63059dd4889bd6a1da14ecb0b0b97303a6f7d2dd1a391a8c2fd8b38bd
|
| 3 |
+
size 40689163
|
Ecapa_tdnn_512_UAMSoftmax_Inter_intra/.DS_Store
ADDED
|
Binary file (6.15 kB). View file
|
|
|
Ecapa_tdnn_512_UAMSoftmax_Inter_intra/config.yaml
ADDED
|
@@ -0,0 +1,87 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
checkpoint: exp/u_cube_plus/ecapa_u_cube_AM_inter_intra/models/model_149.pt
|
| 2 |
+
data_type: shard
|
| 3 |
+
dataloader_args:
|
| 4 |
+
batch_size: 512
|
| 5 |
+
drop_last: true
|
| 6 |
+
num_workers: 12
|
| 7 |
+
pin_memory: false
|
| 8 |
+
prefetch_factor: 8
|
| 9 |
+
dataset_args:
|
| 10 |
+
aug_prob: 0.6
|
| 11 |
+
fbank_args:
|
| 12 |
+
dither: 1.0
|
| 13 |
+
frame_length: 25
|
| 14 |
+
frame_shift: 10
|
| 15 |
+
num_mel_bins: 80
|
| 16 |
+
filter: true
|
| 17 |
+
filter_args:
|
| 18 |
+
max_num_frames: 800
|
| 19 |
+
min_num_frames: 100
|
| 20 |
+
frontend: fbank
|
| 21 |
+
num_frms: 200
|
| 22 |
+
resample_rate: 16000
|
| 23 |
+
sample_num_per_epoch: 0
|
| 24 |
+
shuffle: true
|
| 25 |
+
shuffle_args:
|
| 26 |
+
shuffle_size: 2500
|
| 27 |
+
spec_aug: false
|
| 28 |
+
spec_aug_args:
|
| 29 |
+
max_f: 8
|
| 30 |
+
max_t: 10
|
| 31 |
+
num_f_mask: 1
|
| 32 |
+
num_t_mask: 1
|
| 33 |
+
prob: 0.6
|
| 34 |
+
speed_perturb: true
|
| 35 |
+
enable_amp: false
|
| 36 |
+
exp_dir: exp/u_cube_plus/ecapa_u_cube_AM_inter_intra
|
| 37 |
+
gpus:
|
| 38 |
+
- 2
|
| 39 |
+
- 3
|
| 40 |
+
log_batch_interval: 100
|
| 41 |
+
loss: CrossEntropyLoss
|
| 42 |
+
loss_args: {}
|
| 43 |
+
margin_scheduler: MarginScheduler
|
| 44 |
+
margin_update:
|
| 45 |
+
epoch_iter: 1066
|
| 46 |
+
final_margin: 0.2
|
| 47 |
+
fix_start_epoch: 40
|
| 48 |
+
increase_start_epoch: 20
|
| 49 |
+
increase_type: exp
|
| 50 |
+
initial_margin: 0.0
|
| 51 |
+
update_margin: true
|
| 52 |
+
model: ECAPA_TDNN_GLOB_c512
|
| 53 |
+
model_args:
|
| 54 |
+
embed_dim: 192
|
| 55 |
+
feat_dim: 80
|
| 56 |
+
pooling_func: U_Cube_XI
|
| 57 |
+
model_init: null
|
| 58 |
+
noise_data: data/musan/lmdb
|
| 59 |
+
num_avg: 10
|
| 60 |
+
num_epochs: 150
|
| 61 |
+
optimizer: SGD
|
| 62 |
+
optimizer_args:
|
| 63 |
+
lr: 0.1
|
| 64 |
+
momentum: 0.9
|
| 65 |
+
nesterov: true
|
| 66 |
+
weight_decay: 0.0001
|
| 67 |
+
projection_args:
|
| 68 |
+
do_lm: false
|
| 69 |
+
easy_margin: false
|
| 70 |
+
embed_dim: 192
|
| 71 |
+
num_class: 17982
|
| 72 |
+
project_type: add_margin_uncertainty_inter_intra
|
| 73 |
+
scale: 32
|
| 74 |
+
reverb_data: data/rirs/lmdb
|
| 75 |
+
save_epoch_interval: 10
|
| 76 |
+
scheduler: ExponentialDecrease
|
| 77 |
+
scheduler_args:
|
| 78 |
+
epoch_iter: 1066
|
| 79 |
+
final_lr: 5.0e-05
|
| 80 |
+
initial_lr: 0.1
|
| 81 |
+
num_epochs: 150
|
| 82 |
+
scale_ratio: 16.0
|
| 83 |
+
warm_from_zero: true
|
| 84 |
+
warm_up_epoch: 6
|
| 85 |
+
seed: 42
|
| 86 |
+
train_data: data/vox2_dev/shard.list
|
| 87 |
+
train_label: data/vox2_dev/utt2spk
|
Ecapa_tdnn_512_UAMSoftmax_Inter_intra/models/avg_model.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:5c81468917d59141394a0831f5876b68492e952546791006bb1e4198e380c9fb
|
| 3 |
+
size 40689163
|
RedimNetB2_UAAMSoftmax_Inter_intra/config.yaml
ADDED
|
@@ -0,0 +1,86 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
data_type: shard
|
| 2 |
+
dataloader_args:
|
| 3 |
+
batch_size: 128
|
| 4 |
+
drop_last: true
|
| 5 |
+
num_workers: 12
|
| 6 |
+
pin_memory: false
|
| 7 |
+
prefetch_factor: 4
|
| 8 |
+
dataset_args:
|
| 9 |
+
aug_prob: 0.6
|
| 10 |
+
fbank_args:
|
| 11 |
+
dither: 1.0
|
| 12 |
+
frame_length: 25
|
| 13 |
+
frame_shift: 10
|
| 14 |
+
num_mel_bins: 72
|
| 15 |
+
filter: true
|
| 16 |
+
filter_args:
|
| 17 |
+
max_num_frames: 800
|
| 18 |
+
min_num_frames: 100
|
| 19 |
+
num_frms: 200
|
| 20 |
+
resample_rate: 16000
|
| 21 |
+
sample_num_per_epoch: 0
|
| 22 |
+
shuffle: true
|
| 23 |
+
shuffle_args:
|
| 24 |
+
shuffle_size: 2500
|
| 25 |
+
spec_aug: false
|
| 26 |
+
spec_aug_args:
|
| 27 |
+
max_f: 8
|
| 28 |
+
max_t: 10
|
| 29 |
+
num_f_mask: 1
|
| 30 |
+
num_t_mask: 1
|
| 31 |
+
prob: 0.6
|
| 32 |
+
speed_perturb: true
|
| 33 |
+
enable_amp: false
|
| 34 |
+
exp_dir: exp/redimnet_u_cube_UAAM_inter_intra
|
| 35 |
+
gpus:
|
| 36 |
+
- 2
|
| 37 |
+
- 3
|
| 38 |
+
log_batch_interval: 100
|
| 39 |
+
loss: CrossEntropyLoss
|
| 40 |
+
loss_args: {}
|
| 41 |
+
margin_scheduler: MarginScheduler
|
| 42 |
+
margin_update:
|
| 43 |
+
epoch_iter: 4265
|
| 44 |
+
final_margin: 0.2
|
| 45 |
+
fix_start_epoch: 40
|
| 46 |
+
increase_start_epoch: 20
|
| 47 |
+
increase_type: exp
|
| 48 |
+
initial_margin: 0.0
|
| 49 |
+
update_margin: true
|
| 50 |
+
model: ReDimNetB2
|
| 51 |
+
model_args:
|
| 52 |
+
embed_dim: 192
|
| 53 |
+
feat_dim: 72
|
| 54 |
+
pooling_func: U_Cube_XI
|
| 55 |
+
two_emb_layer: false
|
| 56 |
+
model_init: null
|
| 57 |
+
noise_data: data/musan/lmdb
|
| 58 |
+
num_avg: 10
|
| 59 |
+
num_epochs: 120
|
| 60 |
+
optimizer: SGD
|
| 61 |
+
optimizer_args:
|
| 62 |
+
lr: 0.1
|
| 63 |
+
momentum: 0.9
|
| 64 |
+
nesterov: true
|
| 65 |
+
weight_decay: 2.0e-05
|
| 66 |
+
projection_args:
|
| 67 |
+
do_lm: false
|
| 68 |
+
easy_margin: false
|
| 69 |
+
embed_dim: 192
|
| 70 |
+
num_class: 17982
|
| 71 |
+
project_type: arc_margin_uncertainty_inter_intra
|
| 72 |
+
scale: 32.0
|
| 73 |
+
reverb_data: data/rirs/lmdb
|
| 74 |
+
save_epoch_interval: 5
|
| 75 |
+
scheduler: ExponentialDecrease
|
| 76 |
+
scheduler_args:
|
| 77 |
+
epoch_iter: 4265
|
| 78 |
+
final_lr: 5.0e-05
|
| 79 |
+
initial_lr: 0.1
|
| 80 |
+
num_epochs: 120
|
| 81 |
+
scale_ratio: 4.0
|
| 82 |
+
warm_from_zero: true
|
| 83 |
+
warm_up_epoch: 6
|
| 84 |
+
seed: 42
|
| 85 |
+
train_data: data/vox2_dev/shard.list
|
| 86 |
+
train_label: data/vox2_dev/utt2spk
|
RedimNetB2_UAAMSoftmax_Inter_intra/models/avg_model.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:7107d1ea2bfd2308e3d5f9b46983ea381a568e5008c65f19787cf921116aa1c7
|
| 3 |
+
size 35893643
|
RedimNetB2_USphereface2_Inter_intra/config.yaml
ADDED
|
@@ -0,0 +1,86 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
data_type: shard
|
| 2 |
+
dataloader_args:
|
| 3 |
+
batch_size: 128
|
| 4 |
+
drop_last: true
|
| 5 |
+
num_workers: 12
|
| 6 |
+
pin_memory: false
|
| 7 |
+
prefetch_factor: 4
|
| 8 |
+
dataset_args:
|
| 9 |
+
aug_prob: 0.6
|
| 10 |
+
fbank_args:
|
| 11 |
+
dither: 1.0
|
| 12 |
+
frame_length: 25
|
| 13 |
+
frame_shift: 10
|
| 14 |
+
num_mel_bins: 72
|
| 15 |
+
filter: true
|
| 16 |
+
filter_args:
|
| 17 |
+
max_num_frames: 800
|
| 18 |
+
min_num_frames: 100
|
| 19 |
+
num_frms: 200
|
| 20 |
+
resample_rate: 16000
|
| 21 |
+
sample_num_per_epoch: 0
|
| 22 |
+
shuffle: true
|
| 23 |
+
shuffle_args:
|
| 24 |
+
shuffle_size: 2500
|
| 25 |
+
spec_aug: false
|
| 26 |
+
spec_aug_args:
|
| 27 |
+
max_f: 8
|
| 28 |
+
max_t: 10
|
| 29 |
+
num_f_mask: 1
|
| 30 |
+
num_t_mask: 1
|
| 31 |
+
prob: 0.6
|
| 32 |
+
speed_perturb: true
|
| 33 |
+
enable_amp: false
|
| 34 |
+
exp_dir: exp/redimnet_u_cube_sphereface2_uncertainty_arcguide
|
| 35 |
+
gpus:
|
| 36 |
+
- 2
|
| 37 |
+
- 3
|
| 38 |
+
log_batch_interval: 100
|
| 39 |
+
loss: CrossEntropyLoss
|
| 40 |
+
loss_args: {}
|
| 41 |
+
margin_scheduler: MarginScheduler
|
| 42 |
+
margin_update:
|
| 43 |
+
epoch_iter: 4265
|
| 44 |
+
final_margin: 0.2
|
| 45 |
+
fix_start_epoch: 40
|
| 46 |
+
increase_start_epoch: 20
|
| 47 |
+
increase_type: exp
|
| 48 |
+
initial_margin: 0.0
|
| 49 |
+
update_margin: true
|
| 50 |
+
model: ReDimNetB2
|
| 51 |
+
model_args:
|
| 52 |
+
embed_dim: 192
|
| 53 |
+
feat_dim: 72
|
| 54 |
+
pooling_func: U_Cube_XI
|
| 55 |
+
two_emb_layer: false
|
| 56 |
+
model_init: null
|
| 57 |
+
noise_data: data/musan/lmdb
|
| 58 |
+
num_avg: 10
|
| 59 |
+
num_epochs: 120
|
| 60 |
+
optimizer: SGD
|
| 61 |
+
optimizer_args:
|
| 62 |
+
lr: 0.1
|
| 63 |
+
momentum: 0.9
|
| 64 |
+
nesterov: true
|
| 65 |
+
weight_decay: 2.0e-05
|
| 66 |
+
projection_args:
|
| 67 |
+
do_lm: false
|
| 68 |
+
easy_margin: false
|
| 69 |
+
embed_dim: 192
|
| 70 |
+
num_class: 17982
|
| 71 |
+
project_type: sphereface2_uncertainty_arcguide
|
| 72 |
+
scale: 32.0
|
| 73 |
+
reverb_data: data/rirs/lmdb
|
| 74 |
+
save_epoch_interval: 5
|
| 75 |
+
scheduler: ExponentialDecrease
|
| 76 |
+
scheduler_args:
|
| 77 |
+
epoch_iter: 4265
|
| 78 |
+
final_lr: 5.0e-05
|
| 79 |
+
initial_lr: 0.1
|
| 80 |
+
num_epochs: 120
|
| 81 |
+
scale_ratio: 4.0
|
| 82 |
+
warm_from_zero: true
|
| 83 |
+
warm_up_epoch: 6
|
| 84 |
+
seed: 42
|
| 85 |
+
train_data: data/vox2_dev/shard.list
|
| 86 |
+
train_label: data/vox2_dev/utt2spk
|
RedimNetB2_USphereface2_Inter_intra/models/avg_model.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:26cb1f7afbf8190e992c4ef0e20c050a2f6705e1d24da941b9fa7250c25274f6
|
| 3 |
+
size 35893899
|
ResNet34_UAAMSoftmax_Inter_intra/config.yaml
ADDED
|
@@ -0,0 +1,86 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
data_type: shard
|
| 2 |
+
dataloader_args:
|
| 3 |
+
batch_size: 256
|
| 4 |
+
drop_last: true
|
| 5 |
+
num_workers: 16
|
| 6 |
+
pin_memory: false
|
| 7 |
+
prefetch_factor: 8
|
| 8 |
+
dataset_args:
|
| 9 |
+
aug_prob: 0.6
|
| 10 |
+
fbank_args:
|
| 11 |
+
dither: 1.0
|
| 12 |
+
frame_length: 25
|
| 13 |
+
frame_shift: 10
|
| 14 |
+
num_mel_bins: 80
|
| 15 |
+
filter: true
|
| 16 |
+
filter_args:
|
| 17 |
+
max_num_frames: 800
|
| 18 |
+
min_num_frames: 100
|
| 19 |
+
num_frms: 200
|
| 20 |
+
resample_rate: 16000
|
| 21 |
+
sample_num_per_epoch: 0
|
| 22 |
+
shuffle: true
|
| 23 |
+
shuffle_args:
|
| 24 |
+
shuffle_size: 2500
|
| 25 |
+
spec_aug: false
|
| 26 |
+
spec_aug_args:
|
| 27 |
+
max_f: 8
|
| 28 |
+
max_t: 10
|
| 29 |
+
num_f_mask: 1
|
| 30 |
+
num_t_mask: 1
|
| 31 |
+
prob: 0.6
|
| 32 |
+
speed_perturb: true
|
| 33 |
+
enable_amp: false
|
| 34 |
+
exp_dir: exp/ResNet34_UAAM_inter_intra
|
| 35 |
+
gpus:
|
| 36 |
+
- 0
|
| 37 |
+
- 1
|
| 38 |
+
log_batch_interval: 100
|
| 39 |
+
loss: CrossEntropyLoss
|
| 40 |
+
loss_args: {}
|
| 41 |
+
margin_scheduler: MarginScheduler
|
| 42 |
+
margin_update:
|
| 43 |
+
epoch_iter: 2132
|
| 44 |
+
final_margin: 0.2
|
| 45 |
+
fix_start_epoch: 40
|
| 46 |
+
increase_start_epoch: 20
|
| 47 |
+
increase_type: exp
|
| 48 |
+
initial_margin: 0.0
|
| 49 |
+
update_margin: true
|
| 50 |
+
model: ResNet34
|
| 51 |
+
model_args:
|
| 52 |
+
embed_dim: 256
|
| 53 |
+
feat_dim: 80
|
| 54 |
+
pooling_func: U_Cube_XI
|
| 55 |
+
two_emb_layer: false
|
| 56 |
+
model_init: null
|
| 57 |
+
noise_data: data/musan/lmdb
|
| 58 |
+
num_avg: 10
|
| 59 |
+
num_epochs: 150
|
| 60 |
+
optimizer: SGD
|
| 61 |
+
optimizer_args:
|
| 62 |
+
lr: 0.1
|
| 63 |
+
momentum: 0.9
|
| 64 |
+
nesterov: true
|
| 65 |
+
weight_decay: 0.0001
|
| 66 |
+
projection_args:
|
| 67 |
+
do_lm: false
|
| 68 |
+
easy_margin: false
|
| 69 |
+
embed_dim: 256
|
| 70 |
+
num_class: 17982
|
| 71 |
+
project_type: arc_margin_uncertainty_inter_intra
|
| 72 |
+
scale: 32.0
|
| 73 |
+
reverb_data: data/rirs/lmdb
|
| 74 |
+
save_epoch_interval: 5
|
| 75 |
+
scheduler: ExponentialDecrease
|
| 76 |
+
scheduler_args:
|
| 77 |
+
epoch_iter: 2132
|
| 78 |
+
final_lr: 5.0e-05
|
| 79 |
+
initial_lr: 0.1
|
| 80 |
+
num_epochs: 150
|
| 81 |
+
scale_ratio: 8.0
|
| 82 |
+
warm_from_zero: true
|
| 83 |
+
warm_up_epoch: 6
|
| 84 |
+
seed: 42
|
| 85 |
+
train_data: data/vox2_dev/shard.list
|
| 86 |
+
train_label: data/vox2_dev/utt2spk
|
ResNet34_UAAMSoftmax_Inter_intra/models/avg_model.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:748252ac46c828bff680081187d5e55fb34618c0b26f6939dd0bfa0cc09dfb16
|
| 3 |
+
size 50879179
|