37f56db5af
* [example] added code for training speech command dataset * update kes_model.py * update kes_model.py * format * format * add more comments to explain the new classifier designed for speech command classification task * add copyrigh info * update copyrigh info of classifier.py
53 lines
995 B
YAML
53 lines
995 B
YAML
dataset_conf:
|
|
filter_conf:
|
|
max_length: 2048
|
|
min_length: 0
|
|
resample_conf:
|
|
resample_rate: 16000
|
|
speed_perturb: false
|
|
feature_extraction_conf:
|
|
feature_type: 'mfcc'
|
|
num_ceps: 80
|
|
num_mel_bins: 80
|
|
frame_shift: 10
|
|
frame_length: 25
|
|
dither: 1.0
|
|
feature_dither: 0.0
|
|
spec_aug: true
|
|
spec_aug_conf:
|
|
num_t_mask: 2
|
|
num_f_mask: 2
|
|
max_t: 20
|
|
max_f: 40
|
|
shuffle: true
|
|
shuffle_conf:
|
|
shuffle_size: 1500
|
|
batch_conf:
|
|
batch_size: 100
|
|
|
|
model:
|
|
hidden_dim: 64
|
|
preprocessing:
|
|
type: none
|
|
backbone:
|
|
type: mdtc
|
|
num_stack: 4
|
|
stack_size: 4
|
|
kernel_size: 5
|
|
hidden_dim: 64
|
|
causal: False
|
|
classifier:
|
|
type: global
|
|
dropout: 0.5
|
|
|
|
optim: adam
|
|
optim_conf:
|
|
lr: 0.001
|
|
weight_decay: 0.00005
|
|
|
|
training_config:
|
|
grad_clip: 50
|
|
max_epoch: 100
|
|
log_interval: 10
|
|
criterion: ce
|