Alexgichamba
commited on
Commit
•
2c6fc2d
1
Parent(s):
083001f
initial commit
Browse files- 20epoch.pth +3 -0
- README.md +12 -0
- config.yaml +202 -0
- images/backward_time.png +0 -0
- images/clip.png +0 -0
- images/eer.png +0 -0
- images/forward_time.png +0 -0
- images/gpu_max_cached_mem_GB.png +0 -0
- images/grad_norm.png +0 -0
- images/iter_time.png +0 -0
- images/loss.png +0 -0
- images/loss_scale.png +0 -0
- images/mindcf.png +0 -0
- images/n_trials.png +0 -0
- images/nontrg_mean.png +0 -0
- images/nontrg_std.png +0 -0
- images/optim0_lr0.png +0 -0
- images/optim_step_time.png +0 -0
- images/train_time.png +0 -0
- images/trg_mean.png +0 -0
- images/trg_std.png +0 -0
20epoch.pth
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:05f1f9178ee2d9372623749769bd4cd8ca05fb9425e255211b7216900165eb79
|
3 |
+
size 84970210
|
README.md
ADDED
@@ -0,0 +1,12 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
license: apache-2.0
|
3 |
+
---
|
4 |
+
|
5 |
+
# IDLS24 TEAM33
|
6 |
+
## SKA-TDNN with 4 msSKA blocks
|
7 |
+
|
8 |
+
Results on Vox1-O, after training on VoxCeleb1-dev
|
9 |
+
|
10 |
+
| EER (%) | minDCF|
|
11 |
+
|---------|-------|
|
12 |
+
|2.659| 0.178 |
|
config.yaml
ADDED
@@ -0,0 +1,202 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
config: conf/train_quadms_ska.yaml
|
2 |
+
print_config: false
|
3 |
+
log_level: INFO
|
4 |
+
drop_last_iter: true
|
5 |
+
dry_run: false
|
6 |
+
iterator_type: category
|
7 |
+
valid_iterator_type: sequence
|
8 |
+
output_dir: exp/spk_train_quadms_ska_raw
|
9 |
+
ngpu: 1
|
10 |
+
seed: 0
|
11 |
+
num_workers: 4
|
12 |
+
num_att_plot: 0
|
13 |
+
dist_backend: nccl
|
14 |
+
dist_init_method: env://
|
15 |
+
dist_world_size: null
|
16 |
+
dist_rank: null
|
17 |
+
local_rank: 0
|
18 |
+
dist_master_addr: null
|
19 |
+
dist_master_port: null
|
20 |
+
dist_launcher: null
|
21 |
+
multiprocessing_distributed: false
|
22 |
+
unused_parameters: false
|
23 |
+
sharded_ddp: false
|
24 |
+
cudnn_enabled: true
|
25 |
+
cudnn_benchmark: true
|
26 |
+
cudnn_deterministic: false
|
27 |
+
collect_stats: false
|
28 |
+
write_collected_feats: false
|
29 |
+
max_epoch: 40
|
30 |
+
patience: null
|
31 |
+
val_scheduler_criterion:
|
32 |
+
- valid
|
33 |
+
- loss
|
34 |
+
early_stopping_criterion:
|
35 |
+
- valid
|
36 |
+
- loss
|
37 |
+
- min
|
38 |
+
best_model_criterion:
|
39 |
+
- - valid
|
40 |
+
- eer
|
41 |
+
- min
|
42 |
+
keep_nbest_models: 3
|
43 |
+
nbest_averaging_interval: 0
|
44 |
+
grad_clip: 9999
|
45 |
+
grad_clip_type: 2.0
|
46 |
+
grad_noise: false
|
47 |
+
accum_grad: 1
|
48 |
+
no_forward_run: false
|
49 |
+
resume: true
|
50 |
+
train_dtype: float32
|
51 |
+
use_amp: true
|
52 |
+
log_interval: 100
|
53 |
+
use_matplotlib: true
|
54 |
+
use_tensorboard: true
|
55 |
+
create_graph_in_tensorboard: false
|
56 |
+
use_wandb: false
|
57 |
+
wandb_project: null
|
58 |
+
wandb_id: null
|
59 |
+
wandb_entity: null
|
60 |
+
wandb_name: null
|
61 |
+
wandb_model_log_interval: -1
|
62 |
+
detect_anomaly: false
|
63 |
+
use_adapter: false
|
64 |
+
adapter: lora
|
65 |
+
save_strategy: all
|
66 |
+
adapter_conf: {}
|
67 |
+
pretrain_path: null
|
68 |
+
init_param: []
|
69 |
+
ignore_init_mismatch: false
|
70 |
+
freeze_param: []
|
71 |
+
num_iters_per_epoch: 2239
|
72 |
+
batch_size: 64
|
73 |
+
valid_batch_size: 32
|
74 |
+
batch_bins: 1000000
|
75 |
+
valid_batch_bins: null
|
76 |
+
train_shape_file:
|
77 |
+
- exp/spk_stats_16k/train/speech_shape
|
78 |
+
valid_shape_file:
|
79 |
+
- exp/spk_stats_16k/valid/speech_shape
|
80 |
+
batch_type: folded
|
81 |
+
valid_batch_type: null
|
82 |
+
fold_length:
|
83 |
+
- 120000
|
84 |
+
sort_in_batch: descending
|
85 |
+
shuffle_within_batch: false
|
86 |
+
sort_batch: descending
|
87 |
+
multiple_iterator: false
|
88 |
+
chunk_length: 500
|
89 |
+
chunk_shift_ratio: 0.5
|
90 |
+
num_cache_chunks: 1024
|
91 |
+
chunk_excluded_key_prefixes: []
|
92 |
+
chunk_default_fs: null
|
93 |
+
train_data_path_and_name_and_type:
|
94 |
+
- - dump/raw/dev_vox1/wav.scp
|
95 |
+
- speech
|
96 |
+
- sound
|
97 |
+
- - dump/raw/dev_vox1/utt2spk
|
98 |
+
- spk_labels
|
99 |
+
- text
|
100 |
+
valid_data_path_and_name_and_type:
|
101 |
+
- - dump/raw/test_vox1/trial.scp
|
102 |
+
- speech
|
103 |
+
- sound
|
104 |
+
- - dump/raw/test_vox1/trial2.scp
|
105 |
+
- speech2
|
106 |
+
- sound
|
107 |
+
- - dump/raw/test_vox1/trial_label
|
108 |
+
- spk_labels
|
109 |
+
- text
|
110 |
+
allow_variable_data_keys: false
|
111 |
+
max_cache_size: 0.0
|
112 |
+
max_cache_fd: 32
|
113 |
+
allow_multi_rates: false
|
114 |
+
valid_max_cache_size: null
|
115 |
+
exclude_weight_decay: false
|
116 |
+
exclude_weight_decay_conf: {}
|
117 |
+
optim: adam
|
118 |
+
optim_conf:
|
119 |
+
lr: 0.001
|
120 |
+
weight_decay: 5.0e-05
|
121 |
+
amsgrad: false
|
122 |
+
scheduler: cosineannealingwarmuprestarts
|
123 |
+
scheduler_conf:
|
124 |
+
first_cycle_steps: 11195
|
125 |
+
cycle_mult: 1.0
|
126 |
+
max_lr: 0.001
|
127 |
+
min_lr: 5.0e-06
|
128 |
+
warmup_steps: 1000
|
129 |
+
gamma: 0.75
|
130 |
+
init: null
|
131 |
+
use_preprocessor: true
|
132 |
+
input_size: null
|
133 |
+
target_duration: 3.0
|
134 |
+
spk2utt: dump/raw/dev_vox1/spk2utt
|
135 |
+
spk_num: 1211
|
136 |
+
sample_rate: 16000
|
137 |
+
num_eval: 10
|
138 |
+
rir_scp: ''
|
139 |
+
model_conf:
|
140 |
+
extract_feats_in_collect_stats: false
|
141 |
+
frontend: melspec_torch
|
142 |
+
frontend_conf:
|
143 |
+
preemp: true
|
144 |
+
n_fft: 512
|
145 |
+
log: true
|
146 |
+
win_length: 400
|
147 |
+
hop_length: 160
|
148 |
+
n_mels: 80
|
149 |
+
normalize: mn
|
150 |
+
specaug: null
|
151 |
+
specaug_conf: {}
|
152 |
+
normalize: null
|
153 |
+
normalize_conf: {}
|
154 |
+
encoder: quadms_ska_tdnn
|
155 |
+
encoder_conf:
|
156 |
+
model_scale: 8
|
157 |
+
ndim: 512
|
158 |
+
ska_dim: 128
|
159 |
+
output_size: 1536
|
160 |
+
pooling: chn_attn_stat
|
161 |
+
pooling_conf: {}
|
162 |
+
projector: ska_tdnn
|
163 |
+
projector_conf:
|
164 |
+
output_size: 192
|
165 |
+
preprocessor: spk
|
166 |
+
preprocessor_conf:
|
167 |
+
target_duration: 3.0
|
168 |
+
sample_rate: 16000
|
169 |
+
num_eval: 5
|
170 |
+
noise_apply_prob: 0.5
|
171 |
+
noise_info:
|
172 |
+
- - 1.0
|
173 |
+
- dump/raw/musan_speech.scp
|
174 |
+
- - 4
|
175 |
+
- 7
|
176 |
+
- - 13
|
177 |
+
- 20
|
178 |
+
- - 1.0
|
179 |
+
- dump/raw/musan_noise.scp
|
180 |
+
- - 1
|
181 |
+
- 1
|
182 |
+
- - 0
|
183 |
+
- 15
|
184 |
+
- - 1.0
|
185 |
+
- dump/raw/musan_music.scp
|
186 |
+
- - 1
|
187 |
+
- 1
|
188 |
+
- - 5
|
189 |
+
- 15
|
190 |
+
rir_apply_prob: 0.5
|
191 |
+
rir_scp: dump/raw/rirs.scp
|
192 |
+
loss: aamsoftmax_sc_topk
|
193 |
+
loss_conf:
|
194 |
+
margin: 0.3
|
195 |
+
scale: 30
|
196 |
+
K: 3
|
197 |
+
mp: 0.06
|
198 |
+
k_top: 5
|
199 |
+
required:
|
200 |
+
- output_dir
|
201 |
+
version: '202402'
|
202 |
+
distributed: false
|
images/backward_time.png
ADDED
![]() |
images/clip.png
ADDED
![]() |
images/eer.png
ADDED
![]() |
images/forward_time.png
ADDED
![]() |
images/gpu_max_cached_mem_GB.png
ADDED
![]() |
images/grad_norm.png
ADDED
![]() |
images/iter_time.png
ADDED
![]() |
images/loss.png
ADDED
![]() |
images/loss_scale.png
ADDED
![]() |
images/mindcf.png
ADDED
![]() |
images/n_trials.png
ADDED
![]() |
images/nontrg_mean.png
ADDED
![]() |
images/nontrg_std.png
ADDED
![]() |
images/optim0_lr0.png
ADDED
![]() |
images/optim_step_time.png
ADDED
![]() |
images/train_time.png
ADDED
![]() |
images/trg_mean.png
ADDED
![]() |
images/trg_std.png
ADDED
![]() |