|
|
|
|
###########################################
|
|
|
|
|
# Data #
|
|
|
|
|
###########################################
|
|
|
|
|
augment: True
|
|
|
|
|
batch_size: 32
|
|
|
|
|
num_workers: 2
|
|
|
|
|
num_speakers: 7205 # 1211 vox1, 5994 vox2, 7205 vox1+2, test speakers: 41
|
|
|
|
|
shuffle: True
|
|
|
|
|
skip_prep: False
|
|
|
|
|
split_ratio: 0.9
|
|
|
|
|
chunk_duration: 3.0 # seconds
|
|
|
|
|
random_chunk: True
|
|
|
|
|
verification_file: data/vox1/veri_test2.txt
|
|
|
|
|
|
|
|
|
|
###########################################################
|
|
|
|
|
# FEATURE EXTRACTION SETTING #
|
|
|
|
|
###########################################################
|
|
|
|
|
# currently, we only support fbank
|
|
|
|
|
sr: 16000 # sample rate
|
|
|
|
|
n_mels: 80
|
|
|
|
|
window_size: 400 #25ms, sample rate 16000, 25 * 16000 / 1000 = 400
|
|
|
|
|
hop_size: 160 #10ms, sample rate 16000, 10 * 16000 / 1000 = 160
|
|
|
|
|
|
|
|
|
|
###########################################################
|
|
|
|
|
# MODEL SETTING #
|
|
|
|
|
###########################################################
|
|
|
|
|
# currently, we only support ecapa-tdnn in the ecapa_tdnn.yaml
|
|
|
|
|
# if we want use another model, please choose another configuration yaml file
|
|
|
|
|
model:
|
|
|
|
|
input_size: 80
|
|
|
|
|
channels: [1024, 1024, 1024, 1024, 3072]
|
|
|
|
|
kernel_sizes: [5, 3, 3, 3, 1]
|
|
|
|
|
dilations: [1, 2, 3, 4, 1]
|
|
|
|
|
attention_channels: 128
|
|
|
|
|
lin_neurons: 192
|
|
|
|
|
|
|
|
|
|
###########################################
|
|
|
|
|
# Training #
|
|
|
|
|
###########################################
|
|
|
|
|
seed: 1986 # according from speechbrain configuration
|
|
|
|
|
epochs: 10
|
|
|
|
|
save_interval: 10
|
|
|
|
|
log_interval: 10
|
|
|
|
|
learning_rate: 1e-8
|
|
|
|
|
max_lr: 1e-3
|
|
|
|
|
step_size: 140000
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
###########################################
|
|
|
|
|
# loss #
|
|
|
|
|
###########################################
|
|
|
|
|
margin: 0.2
|
|
|
|
|
scale: 30
|
|
|
|
|
|
|
|
|
|
###########################################
|
|
|
|
|
# Testing #
|
|
|
|
|
###########################################
|
|
|
|
|
global_embedding_norm: True
|
|
|
|
|
embedding_mean_norm: True
|
|
|
|
|
embedding_std_norm: False
|
|
|
|
|
|
|
|
|
|
###########################################
|
|
|
|
|
# score-norm #
|
|
|
|
|
###########################################
|
|
|
|
|
score_norm: s-norm
|
|
|
|
|
cohort_size: 20000 # amount of imposter utterances in normalization cohort
|
|
|
|
|
n_train_snts: 400000 # used for normalization stats
|
|
|
|
|
|
|
|
|
|
|
