Commit
6a035b2
• 1
Parent(s):
acebb36
First model version
Browse files- config.yaml +117 -0
- fbank80_stmn_16k.yaml +28 -0
- feats.yaml +28 -0
- model_ep0070.pth +3 -0
- train.log +71 -0
config.yaml
ADDED
@@ -0,0 +1,117 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
audio_path: data/voxceleb2cat_train_proc_audio_no_sil/wav.scp
|
2 |
+
train_list: data/voxceleb2cat_train_proc_audio_no_sil/lists_xvec/train.scp
|
3 |
+
val_list: data/voxceleb2cat_train_proc_audio_no_sil/lists_xvec/val.scp
|
4 |
+
class_file: data/voxceleb2cat_train_proc_audio_no_sil/lists_xvec/class2int
|
5 |
+
time_durs_file: data/voxceleb2cat_train_proc_audio_no_sil/utt2dur
|
6 |
+
min_chunk_length: 4.0
|
7 |
+
max_chunk_length: 4.0
|
8 |
+
return_fullseqs: false
|
9 |
+
wav_scale: 32767
|
10 |
+
batch_size: 512
|
11 |
+
var_batch_size: false
|
12 |
+
iters_per_epoch: 6.0
|
13 |
+
num_egs_per_class: 1
|
14 |
+
num_egs_per_utt: 1
|
15 |
+
train_aug_cfg: conf/reverb_noise_aug.yaml
|
16 |
+
val_aug_cfg: conf/reverb_noise_aug.yaml
|
17 |
+
num_workers: 8
|
18 |
+
feats: fbank80_stmn_16k.yaml
|
19 |
+
pool_net:
|
20 |
+
pool_type: mean+stddev
|
21 |
+
inner_feats: 0
|
22 |
+
num_comp: 8
|
23 |
+
dist_pow: 2
|
24 |
+
wo_bias: false
|
25 |
+
num_heads: 4
|
26 |
+
d_k: 256
|
27 |
+
d_v: 256
|
28 |
+
bin_attn: false
|
29 |
+
embed_dim: 256
|
30 |
+
num_embed_layers: 1
|
31 |
+
hid_act: relu6
|
32 |
+
loss_type: arc-softmax
|
33 |
+
s: 30.0
|
34 |
+
margin: 0.3
|
35 |
+
margin_warmup_epochs: 20.0
|
36 |
+
num_subcenters: 2
|
37 |
+
wo_norm: false
|
38 |
+
norm_after: false
|
39 |
+
dropout_rate: 0.0
|
40 |
+
in_feats: 80
|
41 |
+
resnet_type: lresnet34
|
42 |
+
in_channels: 1
|
43 |
+
conv_channels: 64
|
44 |
+
base_channels: 64
|
45 |
+
in_kernel_size: 3
|
46 |
+
in_stride: 1
|
47 |
+
groups: 1
|
48 |
+
in_norm: false
|
49 |
+
no_maxpool: true
|
50 |
+
zero_init_residual: false
|
51 |
+
se_r: 16
|
52 |
+
res2net_scale: 4
|
53 |
+
res2net_width_factor: 1
|
54 |
+
optim:
|
55 |
+
opt_type: adam
|
56 |
+
lr: 0.05
|
57 |
+
momentum: 0.6
|
58 |
+
beta1: 0.9
|
59 |
+
beta2: 0.95
|
60 |
+
rho: 0.9
|
61 |
+
eps: 1.0e-08
|
62 |
+
weight_decay: 1.0e-05
|
63 |
+
amsgrad: true
|
64 |
+
nesterov: false
|
65 |
+
lambd: 0.0001
|
66 |
+
asgd_alpha: 0.75
|
67 |
+
t0: 1000000.0
|
68 |
+
rmsprop_alpha: 0.99
|
69 |
+
centered: false
|
70 |
+
lr_decay: 1.0e-06
|
71 |
+
init_acc_val: 0
|
72 |
+
max_iter: 20
|
73 |
+
lrsched:
|
74 |
+
lrsch_type: exp_lr
|
75 |
+
decay_rate: 0.5
|
76 |
+
decay_steps: 8000
|
77 |
+
power: 0.5
|
78 |
+
hold_steps: 40000
|
79 |
+
t: 10
|
80 |
+
t_mul: 1
|
81 |
+
gamma: 0.01
|
82 |
+
warm_restarts: false
|
83 |
+
monitor: val_loss
|
84 |
+
mode: min
|
85 |
+
factor: 0.1
|
86 |
+
patience: 10
|
87 |
+
threshold: 0.0001
|
88 |
+
threshold_mode: rel
|
89 |
+
cooldown: 0
|
90 |
+
eps: 1.0e-08
|
91 |
+
min_lr: 1.0e-05
|
92 |
+
warmup_steps: 1000
|
93 |
+
update_lr_on_opt_step: true
|
94 |
+
grad_acc_steps: 1
|
95 |
+
epochs: 70
|
96 |
+
log_interval: 100
|
97 |
+
use_tensorboard: false
|
98 |
+
use_wandb: false
|
99 |
+
wandb:
|
100 |
+
mode: online
|
101 |
+
ddp_type: ddp
|
102 |
+
use_amp: false
|
103 |
+
cpu_offload: false
|
104 |
+
grad_clip: 0
|
105 |
+
grad_clip_norm: 2
|
106 |
+
swa_start: 0
|
107 |
+
swa_lr: 0.001
|
108 |
+
swa_anneal_epochs: 10
|
109 |
+
exp_path: exp/xvector_nnets/fbank80_stmn_lresnet34_e256_arcs30m0.3_do0_adam_lr0.05_b512.v1
|
110 |
+
num_gpus: 4
|
111 |
+
node_id: 0
|
112 |
+
num_nodes: 1
|
113 |
+
master_addr: localhost
|
114 |
+
master_port: '1234'
|
115 |
+
seed: 1123581321
|
116 |
+
resume: false
|
117 |
+
verbose: 1
|
fbank80_stmn_16k.yaml
ADDED
@@ -0,0 +1,28 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
audio_feats:
|
2 |
+
sample_frequency: 16000
|
3 |
+
frame_length: 25
|
4 |
+
frame_shift: 10
|
5 |
+
fft_length: 512
|
6 |
+
remove_dc_offset: true
|
7 |
+
preemphasis_coeff: 0.97
|
8 |
+
window_type: povey
|
9 |
+
use_fft_mag: false
|
10 |
+
dither: 1
|
11 |
+
fb_type: mel_kaldi
|
12 |
+
num_filters: 80
|
13 |
+
low_freq: 20.0
|
14 |
+
high_freq: 7600.0
|
15 |
+
norm_filters: false
|
16 |
+
num_ceps: 13
|
17 |
+
snip_edges: false
|
18 |
+
energy_floor: 0
|
19 |
+
raw_energy: true
|
20 |
+
use_energy: false
|
21 |
+
cepstral_lifter: 22
|
22 |
+
audio_feat: logfb
|
23 |
+
mvn:
|
24 |
+
no_norm_mean: false
|
25 |
+
norm_var: false
|
26 |
+
left_context: 150
|
27 |
+
right_context: 150
|
28 |
+
context: 150
|
feats.yaml
ADDED
@@ -0,0 +1,28 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
audio_feats:
|
2 |
+
sample_frequency: 16000
|
3 |
+
frame_length: 25
|
4 |
+
frame_shift: 10
|
5 |
+
fft_length: 512
|
6 |
+
remove_dc_offset: true
|
7 |
+
preemphasis_coeff: 0.97
|
8 |
+
window_type: povey
|
9 |
+
use_fft_mag: false
|
10 |
+
dither: 1
|
11 |
+
fb_type: mel_kaldi
|
12 |
+
num_filters: 80
|
13 |
+
low_freq: 20.0
|
14 |
+
high_freq: 7600.0
|
15 |
+
norm_filters: false
|
16 |
+
num_ceps: 13
|
17 |
+
snip_edges: false
|
18 |
+
energy_floor: 0
|
19 |
+
raw_energy: true
|
20 |
+
use_energy: false
|
21 |
+
cepstral_lifter: 22
|
22 |
+
audio_feat: logfb
|
23 |
+
mvn:
|
24 |
+
no_norm_mean: false
|
25 |
+
norm_var: false
|
26 |
+
left_context: 150
|
27 |
+
right_context: 150
|
28 |
+
context: 150
|
model_ep0070.pth
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:092c8312cfb41c590adc654f06fe05fd73135c3396b4e3bf65f085a521eeac3d
|
3 |
+
size 56592479
|
train.log
ADDED
@@ -0,0 +1,71 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
epoch,train_loss,train_acc,lr,val_loss,val_acc
|
2 |
+
1,4.831538677215576,0.2949525713920593,0.05,3.324049949645996,0.4054264724254608
|
3 |
+
2,1.7885594367980957,0.6740760207176208,0.05,2.5202794075012207,0.5397918820381165
|
4 |
+
3,1.5020267963409424,0.7304190993309021,0.05,2.4665963649749756,0.5605702996253967
|
5 |
+
4,1.4886183738708496,0.7337057590484619,0.05,2.176955461502075,0.6156345009803772
|
6 |
+
5,1.5453656911849976,0.7233652472496033,0.05,2.0865378379821777,0.6330476999282837
|
7 |
+
6,1.6363445520401,0.706717312335968,0.05,1.8307287693023682,0.6822760105133057
|
8 |
+
7,1.7530429363250732,0.6845660209655762,0.05,1.9427416324615479,0.672086775302887
|
9 |
+
8,1.8867443799972534,0.6606240272521973,0.05,1.7581342458724976,0.6943005323410034
|
10 |
+
9,2.038376808166504,0.6325106620788574,0.05,1.6445910930633545,0.7192911505699158
|
11 |
+
10,2.2235965728759766,0.5997515916824341,0.05,1.668384075164795,0.7167733907699585
|
12 |
+
11,2.4104247093200684,0.5693153142929077,0.05,1.7388219833374023,0.6979952454566956
|
13 |
+
12,2.62192440032959,0.5344089865684509,0.05,1.590958595275879,0.7271037101745605
|
14 |
+
13,2.830293893814087,0.5015796422958374,0.05,1.5722118616104126,0.7273390889167786
|
15 |
+
14,3.0677943229675293,0.4668572247028351,0.05,1.4923597574234009,0.7453406453132629
|
16 |
+
15,3.3032259941101074,0.43319013714790344,0.05,1.6153490543365479,0.7227032780647278
|
17 |
+
16,3.567964553833008,0.39844244718551636,0.05,1.5561275482177734,0.7292451858520508
|
18 |
+
17,3.8298768997192383,0.3645962178707123,0.05,1.4156708717346191,0.7608715891838074
|
19 |
+
18,4.101251125335693,0.3335207402706146,0.05,1.5563970804214478,0.7342573404312134
|
20 |
+
19,4.414449214935303,0.2981777787208557,0.05,1.393101453781128,0.7610835433006287
|
21 |
+
20,4.711869716644287,0.26825350522994995,0.05,1.3954812288284302,0.7597658038139343
|
22 |
+
21,5.046350479125977,0.2379523366689682,0.05,1.4474290609359741,0.7544475197792053
|
23 |
+
22,4.99770975112915,0.2434740662574768,0.05,1.3176016807556152,0.769649088382721
|
24 |
+
23,4.953365325927734,0.24839191138744354,0.05,1.5722708702087402,0.730421781539917
|
25 |
+
24,4.9378886222839355,0.25038155913352966,0.05,1.489615797996521,0.7417168617248535
|
26 |
+
25,4.901826858520508,0.25338420271873474,0.05,1.4267014265060425,0.7554357051849365
|
27 |
+
26,4.821268081665039,0.26222503185272217,0.043826488625414495,1.3263622522354126,0.7753200531005859
|
28 |
+
27,4.681831359863281,0.2765805721282959,0.03816309281110629,1.2655586004257202,0.7820736169815063
|
29 |
+
28,4.558086395263672,0.28974246978759766,0.033231538701563916,1.1349343061447144,0.8061699271202087
|
30 |
+
29,4.446705341339111,0.3025340139865875,0.02893725542475306,1.1010875701904297,0.8165005445480347
|
31 |
+
30,4.349989414215088,0.31403130292892456,0.025197892852249822,1.0429977178573608,0.8255365490913391
|
32 |
+
31,4.252237319946289,0.32567647099494934,0.021941742396562505,1.0128703117370605,0.8304075598716736
|
33 |
+
32,4.169577598571777,0.33487826585769653,0.019106361877958486,0.9141005873680115,0.8509976863861084
|
34 |
+
33,4.070428848266602,0.34717458486557007,0.016637378090296792,1.0088523626327515,0.8316547870635986
|
35 |
+
34,4.005584239959717,0.3554088771343231,0.014487444103045748,0.9605793356895447,0.8383141160011292
|
36 |
+
35,3.9355697631835938,0.36390602588653564,0.012615331303992195,0.9302313923835754,0.8477269411087036
|
37 |
+
36,3.8655645847320557,0.37288570404052734,0.010985138771029147,0.8441817760467529,0.8636106252670288
|
38 |
+
37,3.8035202026367188,0.3816739320755005,0.009565604811391673,0.8162680864334106,0.8665757179260254
|
39 |
+
38,3.7418298721313477,0.38854366540908813,0.008329507465944125,0.8164356350898743,0.8674933910369873
|
40 |
+
39,3.6942782402038574,0.3958593010902405,0.007253142482176712,0.7882479429244995,0.870152473449707
|
41 |
+
40,3.647254705429077,0.4026658236980438,0.006315868745162784,0.77254319190979,0.8747411370277405
|
42 |
+
41,3.598754644393921,0.410393625497818,0.005499712449348278,0.7799084186553955,0.8734704852104187
|
43 |
+
42,3.55338454246521,0.41485852003097534,0.004789022420499469,0.7606723308563232,0.8802004456520081
|
44 |
+
43,3.516636848449707,0.41994646191596985,0.00417016997802574,0.7473565936088562,0.8799886703491211
|
45 |
+
44,3.4718680381774902,0.4265800416469574,0.003631287582030883,0.7510011196136475,0.8802005052566528
|
46 |
+
45,3.4465198516845703,0.4300559461116791,0.0031620412532091524,0.7157604098320007,0.8849068284034729
|
47 |
+
46,3.41135573387146,0.4342111647129059,0.0027534324013535195,0.7078049182891846,0.8888600468635559
|
48 |
+
47,3.383302927017212,0.439024955034256,0.002397625262203351,0.7082321047782898,0.8867660164833069
|
49 |
+
48,3.3596179485321045,0.44231608510017395,0.0020877966334418865,0.6603009700775146,0.8974964022636414
|
50 |
+
49,3.335097551345825,0.44599854946136475,0.0018180050282776768,0.68424391746521,0.891142725944519
|
51 |
+
50,3.3127121925354004,0.4490085244178772,0.0015830767374091165,0.6633255481719971,0.8959667682647705
|
52 |
+
51,3.302381992340088,0.45040014386177063,0.0013785066144179616,0.6618404388427734,0.8955665826797485
|
53 |
+
52,3.2817506790161133,0.452764093875885,0.0012003716819843453,0.660285234451294,0.8946017622947693
|
54 |
+
53,3.2702953815460205,0.45552584528923035,0.0010452559021766502,0.6358559727668762,0.8977317214012146
|
55 |
+
54,3.259110689163208,0.457537442445755,0.0009101846681596171,0.6480079293251038,0.8973550796508789
|
56 |
+
55,3.241170883178711,0.4600141942501068,0.0007925677610886383,0.6479377150535583,0.8979431986808777
|
57 |
+
56,3.2366254329681396,0.46083956956863403,0.000690149678292425,0.6325076818466187,0.9003907442092896
|
58 |
+
57,3.224062204360962,0.46273162961006165,0.000600966380203129,0.6679848432540894,0.8942252993583679
|
59 |
+
58,3.207883358001709,0.46397873759269714,0.0005233076265833215,0.6351966857910156,0.9016614556312561
|
60 |
+
59,3.205732822418213,0.4645356833934784,0.00045568417978341215,0.6340398788452148,0.9011436700820923
|
61 |
+
60,3.2012877464294434,0.46493542194366455,0.00039679924609663453,0.6334475874900818,0.9017319083213806
|
62 |
+
61,3.194880247116089,0.46584776043891907,0.00034552360755138315,0.6428933143615723,0.8985550999641418
|
63 |
+
62,3.1945083141326904,0.4667612612247467,0.0003008739672510551,0.6307220458984375,0.8999906182289124
|
64 |
+
63,3.1938371658325195,0.4674670100212097,0.0002619940929967481,0.6613357663154602,0.8965549468994141
|
65 |
+
64,3.180534601211548,0.4693319499492645,0.00022813839758995628,0.6404596567153931,0.8968846201896667
|
66 |
+
65,3.173895835876465,0.4689936637878418,0.00019865764093986266,0.6339607834815979,0.899825930595398
|
67 |
+
66,3.172269105911255,0.46927976608276367,0.000172986479789007,0.638012170791626,0.9017319083213806
|
68 |
+
67,3.1737780570983887,0.4699510633945465,0.0001506326263023084,0.6294083595275879,0.9010729789733887
|
69 |
+
68,3.1733322143554688,0.46895837783813477,0.00013116740761709425,0.6304994821548462,0.9010024666786194
|
70 |
+
69,3.167774200439453,0.47052833437919617,0.00011421754531757306,0.6343788504600525,0.9002963900566101
|
71 |
+
70,3.161440849304199,0.4707949459552765,9.945799719130592e-05,0.627429723739624,0.902579128742218
|