# Feature extractor CNN
CNN_arch = {
'input_dim': wlen,
'fs': fs,
'cnn_N_filt': cnn_N_filt,
'cnn_len_filt': cnn_len_filt,
'cnn_max_pool_len': cnn_max_pool_len,
'cnn_use_laynorm_inp': cnn_use_laynorm_inp,
'cnn_use_batchnorm_inp': cnn_use_batchnorm_inp,
'cnn_use_laynorm': cnn_use_laynorm,
'cnn_use_batchnorm': cnn_use_batchnorm,
'cnn_act': cnn_act,
'cnn_drop': cnn_drop
}
CNN_net = CNN(CNN_arch)
CNN_net.to(device)
DNN1_arch = {
'input_dim': CNN_net.out_dim,
'fc_lay': fc_lay,
'fc_drop': fc_drop,
'fc_use_batchnorm': fc_use_batchnorm,
'fc_use_laynorm': fc_use_laynorm,
'fc_use_laynorm_inp': fc_use_laynorm_inp,
'fc_use_batchnorm_inp': fc_use_batchnorm_inp,
'fc_act': fc_act
}
DNN1_net = MLP(DNN1_arch)
DNN1_net.to(device)
# Feature extractor CNN
CNN_arch = {'input_dim': wlen,
'fs': fs,
'cnn_N_filt': cnn_N_filt,
'cnn_len_filt': cnn_len_filt,
'cnn_max_pool_len':cnn_max_pool_len,
'cnn_use_laynorm_inp': cnn_use_laynorm_inp,
'cnn_use_batchnorm_inp': cnn_use_batchnorm_inp,
'cnn_use_laynorm':cnn_use_laynorm,
'cnn_use_batchnorm':cnn_use_batchnorm,
'cnn_act': cnn_act,
'cnn_drop':cnn_drop,
}
CNN_net=CNN(CNN_arch)
CNN_net.cuda()
# Loading label dictionary
lab_dict=np.load(class_dict_file).item()
print(CNN_net.out_dim)
DNN1_arch = {'input_dim': CNN_net.out_dim,
'fc_lay': fc_lay,
'fc_drop': fc_drop,
'fc_use_batchnorm': fc_use_batchnorm,
'fc_use_laynorm': fc_use_laynorm,
'fc_use_laynorm_inp': fc_use_laynorm_inp,
'fc_use_batchnorm_inp':fc_use_batchnorm_inp,
'fc_act': fc_act,
# Feature extractor CNN
CNN_arch = {
'input_dim': wlen,
'fs': fs,
'cnn_N_filt': cnn_N_filt,
'cnn_len_filt': cnn_len_filt,
'cnn_max_pool_len': cnn_max_pool_len,
'cnn_use_laynorm_inp': cnn_use_laynorm_inp,
'cnn_use_batchnorm_inp': cnn_use_batchnorm_inp,
'cnn_use_laynorm': cnn_use_laynorm,
'cnn_use_batchnorm': cnn_use_batchnorm,
'cnn_act': cnn_act,
'cnn_drop': cnn_drop,
}
CNN_net = CNN(CNN_arch)
CNN_net.cuda()
# Loading label dictionary
lab_dict = np.load(class_dict_file).item()
DNN1_arch = {
'input_dim': CNN_net.out_dim,
'fc_lay': fc_lay,
'fc_drop': fc_drop,
'fc_use_batchnorm': fc_use_batchnorm,
'fc_use_laynorm': fc_use_laynorm,
'fc_use_laynorm_inp': fc_use_laynorm_inp,
'fc_use_batchnorm_inp': fc_use_batchnorm_inp,
'fc_act': fc_act,
}
# Feature extractor CNN
CNN_arch = {
'input_dim': wlen,
'fs': fs,
'cnn_N_filt': cnn_N_filt,
'cnn_len_filt': cnn_len_filt,
'cnn_max_pool_len': cnn_max_pool_len,
'cnn_use_laynorm_inp': cnn_use_laynorm_inp,
'cnn_use_batchnorm_inp': cnn_use_batchnorm_inp,
'cnn_use_laynorm': cnn_use_laynorm,
'cnn_use_batchnorm': cnn_use_batchnorm,
'cnn_act': cnn_act,
'cnn_drop': cnn_drop,
}
CNN_net = CNN(CNN_arch)
CNN_net.cuda()
# Loading label dictionary
lab_dict = np.load(class_dict_file).item()
DNN1_arch = {
'input_dim': CNN_net.out_dim,
'fc_lay': fc_lay,
'fc_drop': fc_drop,
'fc_use_batchnorm': fc_use_batchnorm,
'fc_use_laynorm': fc_use_laynorm,
'fc_use_laynorm_inp': fc_use_laynorm_inp,
'fc_use_batchnorm_inp': fc_use_batchnorm_inp,
'fc_act': fc_act,
}
# Feature extractor CNN(SincNet)
CNN_arch = {
'input_dim': wlen,
'fs': fs,
'cnn_N_filt': cnn_N_filt,
'cnn_len_filt': cnn_len_filt,
'cnn_max_pool_len': cnn_max_pool_len,
'cnn_use_laynorm_inp': cnn_use_laynorm_inp,
'cnn_use_batchnorm_inp': cnn_use_batchnorm_inp,
'cnn_use_laynorm': cnn_use_laynorm,
'cnn_use_batchnorm': cnn_use_batchnorm,
'cnn_act': cnn_act,
'cnn_drop': cnn_drop,
}
CNN_net = CNN(CNN_arch)
CNN_net.to(device)
# 建立注意力机制
# print(CNN_net.out_dim) 6420
AttentionModule = DoubleMHA(CNN_net.out_dim, 20) # 8 16 32的头数
# 修改这里保证pooling中assert self.encoder_size % heads_number == 0 # d_model 可以通过
# Loading label dictionary
lab_dict = np.load(class_dict_file, allow_pickle=True).item()
# print(CNN_net.out_dim) 6420
DNN1_arch = {
'input_dim':
321, # CNN_net.out_dim where 321 means after attention machine dim
'fc_lay': fc_lay,
else:
if architecture in ['SincNet_raw', 'CNN_raw']:
CNN_arch = {'input_dim': wlen,
'fs': fs,
'cnn_N_filt': cnn_N_filt,
'cnn_len_filt': cnn_len_filt,
'cnn_max_pool_len':cnn_max_pool_len,
'cnn_use_laynorm_inp': cnn_use_laynorm_inp,
'cnn_use_batchnorm_inp': cnn_use_batchnorm_inp,
'cnn_use_laynorm':cnn_use_laynorm,
'cnn_use_batchnorm':cnn_use_batchnorm,
'cnn_act': cnn_act,
'cnn_drop':cnn_drop,
}
if architecture == 'SincNet_raw':
CNN_net = SincNet(CNN_arch)
else:
CNN_net = ConvNet(CNN_arch)
MLP_after = {'input_dim': CNN_net.out_dim,
'fc_lay': fc_lay,
'fc_drop': fc_drop,
'fc_use_batchnorm': fc_use_batchnorm,
'fc_use_laynorm': fc_use_laynorm,
'fc_use_laynorm_inp': fc_use_laynorm_inp,
'fc_use_batchnorm_inp':fc_use_batchnorm_inp,
'fc_act': fc_act,
}
model = FunTimesCNN(MLP_before, MLP_after, CNN_arch, use_sinc_net=architecture=='SincNet_raw')
elif architecture == 'CNN_features':
try:
os.stat(output_folder)
except:
os.mkdir(output_folder)
# setting seed
torch.manual_seed(seed)
np.random.seed(seed)
# Loading label dictionary
lab_dict = np.load(class_dict_file, allow_pickle=True).item()
# 读取模型
checkpoint = torch.load('exp/SincNet_TIMIT/model_raw')
CNN_net = CNN({})
CNN_net.load_state_dict(checkpoint['CNN_model_par'])
DNN1_net = MLP({})
DNN1_net.load_state_dict(checkpoint['DNN1_model_par'])
DNN2_net = MLP({})
DNN2_net.load_state_dict(checkpoint['DNN2_model_par'])
eer = 0
for i in range(N_batches):
# 将test信号预处理batch TODO data—folder是test数据集
[inp, lab] = create_batches_rnd(batch_size, data_folder, wav_lst_te,
snt_te, wlen, lab_dict, 0.2)
pout = DNN2_net(DNN1_net(CNN_net(inp)))
pred = torch.max(pout, dim=1)[1] # 寻找最大那个就是预测的谁
fpr, tpr, thresholds = roc_curve(lab, pred, pos_label=1)
