def __init__(self, args, adj, inp, pretrained=True, model_pth=None, fixed=True):
super(gcnNet, self).__init__()
self.spec_augmenter = SpecAugmentation(time_drop_width=64, time_stripes_num=2,
freq_drop_width=8, freq_stripes_num=2)
self.bn0 = nn.BatchNorm2d(64)
self.conv_block1 = ConvBlock(1,64)
self.conv_block2 = ConvBlock(64,128)
self.conv_block3 = ConvBlock(128,256)
self.conv_block4 = ConvBlock(256,512)
self.fc1 = nn.Linear(512, 512, bias=True)
# self.fc_audioset = nn.Linear(512, 527, bias=True)
if pretrained==True:
self.init_model(model_pth=model_pth)
if fixed==True:
for p in self.parameters():
p.requires_grad=False
self.gcn = GCNnet(num_classes=527, in_channel=300, t=0.3, adj_file=adj)
# self.alpha = nn.Parameter(torch.cuda.FloatTensor([.5, .5]))
with open(inp, 'rb') as f:
self.inp = pickle.load(f)
def __init__(self, sample_rate, window_size, hop_size, mel_bins, fmin,
fmax, classes_num):
super(Cnn_9layers_FrameAtt, self).__init__()
window = 'hann'
center = True
pad_mode = 'reflect'
ref = 1.0
amin = 1e-10
top_db = None
# Spectrogram extractor
self.spectrogram_extractor = Spectrogram(n_fft=window_size, hop_length=hop_size,
win_length=window_size, window=window, center=center, pad_mode=pad_mode,
freeze_parameters=True)
# Logmel feature extractor
self.logmel_extractor = LogmelFilterBank(sr=sample_rate, n_fft=window_size,
n_mels=mel_bins, fmin=fmin, fmax=fmax, ref=ref, amin=amin, top_db=top_db,
freeze_parameters=True)
# Spec augmenter
self.spec_augmenter = SpecAugmentation(time_drop_width=64, time_stripes_num=2,
freq_drop_width=8, freq_stripes_num=2)
self.bn0 = nn.BatchNorm2d(64)
self.conv_block1 = ConvBlock(in_channels=1, out_channels=64)
self.conv_block2 = ConvBlock(in_channels=64, out_channels=128)
self.conv_block3 = ConvBlock(in_channels=128, out_channels=256)
self.conv_block4 = ConvBlock(in_channels=256, out_channels=512)
self.att_block = AttBlock(n_in=512, n_out=17, activation='sigmoid')
self.init_weights()
def __init__(self, sample_rate, window_size, hop_size, mel_bins, fmin,
fmax, classes_num):
super(Cnn_9layers_Gru_FrameAvg, self).__init__()
window = 'hann'
center = True
pad_mode = 'reflect'
ref = 1.0
amin = 1e-10
top_db = None
# Spectrogram extractor
self.spectrogram_extractor = Spectrogram(n_fft=window_size,
hop_length=hop_size,
win_length=window_size,
window=window,
center=center,
pad_mode=pad_mode,
freeze_parameters=True)
# Logmel feature extractor
self.logmel_extractor = LogmelFilterBank(sr=sample_rate,
n_fft=window_size,
n_mels=mel_bins,
fmin=fmin,
fmax=fmax,
ref=ref,
amin=amin,
top_db=top_db,
freeze_parameters=True)
# Spec augmenter
self.spec_augmenter = SpecAugmentation(time_drop_width=64,
time_stripes_num=2,
freq_drop_width=8,
freq_stripes_num=2)
self.bn0 = nn.BatchNorm2d(64)
self.conv_block1 = ConvBlock(in_channels=1, out_channels=64)
self.conv_block2 = ConvBlock(in_channels=64, out_channels=128)
self.conv_block3 = ConvBlock(in_channels=128, out_channels=256)
self.conv_block4 = ConvBlock(in_channels=256, out_channels=512)
self.gru = nn.GRU(input_size=512,
hidden_size=256,
num_layers=1,
bias=True,
batch_first=True,
bidirectional=True)
self.fc = nn.Linear(512, classes_num, bias=True)
self.init_weights()
def __init__(self, args):
super(CNN10, self).__init__()
self.spec_augmenter = SpecAugmentation(time_drop_width=64, time_stripes_num=2,
freq_drop_width=8, freq_stripes_num=2)
self.bn0 = nn.BatchNorm2d(64)
self.conv_block1 = ConvBlock(1,64)
self.conv_block2 = ConvBlock(64,128)
self.conv_block3 = ConvBlock(128,256)
self.conv_block4 = ConvBlock(256,512)
self.fc1 = nn.Linear(512, 512, bias=True)
self.fc_audioset = nn.Linear(512, 527, bias=True)
self.init_weight()
def __init__(self, sample_rate, window_size, hop_size, mel_bins, fmin,
fmax, classes_num):
super(Cnn_9layers_Transformer_FrameAvg, self).__init__()
window = 'hann'
center = True
pad_mode = 'reflect'
ref = 1.0
amin = 1e-10
top_db = None
# Spectrogram extractor
self.spectrogram_extractor = Spectrogram(n_fft=window_size, hop_length=hop_size,
win_length=window_size, window=window, center=center, pad_mode=pad_mode,
freeze_parameters=True)
# Logmel feature extractor
self.logmel_extractor = LogmelFilterBank(sr=sample_rate, n_fft=window_size,
n_mels=mel_bins, fmin=fmin, fmax=fmax, ref=ref, amin=amin, top_db=top_db,
freeze_parameters=True)
# Spec augmenter
self.spec_augmenter = SpecAugmentation(time_drop_width=64, time_stripes_num=2,
freq_drop_width=8, freq_stripes_num=2)
self.bn0 = nn.BatchNorm2d(64)
self.conv_block1 = ConvBlock(in_channels=1, out_channels=64)
self.conv_block2 = ConvBlock(in_channels=64, out_channels=128)
self.conv_block3 = ConvBlock(in_channels=128, out_channels=256)
self.conv_block4 = ConvBlock(in_channels=256, out_channels=512)
n_head = 8
n_hid = 512
d_k = 64
d_v = 64
dropout = 0.2
self.multihead = MultiHead(n_head, n_hid, d_k, d_v, dropout)
self.fc = nn.Linear(512, classes_num, bias=True)
self.init_weights()
def __init__(self, classes_num, time_steps, freq_bins, spec_aug=False):
super(Cnn6, self).__init__()
# Spec augmenter
self.spec_aug = spec_aug
if self.spec_aug:
self.spec_augmenter = SpecAugmentation(
time_drop_width=64,
time_stripes_num=2,
freq_drop_width=8,
freq_stripes_num=2,
)
self.conv_block1 = ConvBlock5x5(in_channels=1, out_channels=64)
self.conv_block2 = ConvBlock5x5(in_channels=64, out_channels=128)
self.conv_block3 = ConvBlock5x5(in_channels=128, out_channels=256)
self.conv_block4 = ConvBlock5x5(in_channels=256, out_channels=512)
self.fc1 = nn.Linear(512, 512, bias=True)
self.fc_out = nn.Linear(512, classes_num, bias=True)
self.init_weight()
def __init__(self,
input_dim,
hidden_dim,
value_size=128,
key_size=128,
pBLSTM_time_reductions=[2],
use_spec_augment=False,
use_conv_blocks_in_encoder=False):
super(Encoder, self).__init__()
self.use_dropout = True
self.use_spec_augment = use_spec_augment
print("Encoder, using spec augment:", use_spec_augment)
self.dropout_layer = nn.Dropout(p=0.1)
print("Encoder, using dropout:", self.use_dropout, str(0.1))
self.spec_augmenter = SpecAugmentation(time_drop_width=64,
time_stripes_num=2,
freq_drop_width=4,
freq_stripes_num=2)
self.use_conv_blocks_in_encoder = use_conv_blocks_in_encoder
if self.use_conv_blocks_in_encoder:
self.conv_block = nn.Sequential(
nn.Conv2d(1, 64, 3, padding=1),
nn.BatchNorm2d(64),
nn.ReLU(),
nn.MaxPool2d(2, 2),
nn.Conv2d(64, 64, 3, padding=1),
nn.BatchNorm2d(64),
nn.ReLU(),
# nn.Conv2d(64, 1, 3, padding=1),
# nn.ReLU(),
nn.MaxPool2d(2, 2))
self.lstm = nn.LSTM(input_size=input_dim,
hidden_size=hidden_dim,
num_layers=1,
bidirectional=True)
nb_pBLSTM_layers = len(pBLSTM_time_reductions)
self.nb_pBLSTM_layers = nb_pBLSTM_layers
reduction_time_factor = pBLSTM_time_reductions[0]
self.pblstm1 = pBLSTM(input_dim=hidden_dim * 2 * reduction_time_factor,
hidden_dim=hidden_dim,
reduction_time_factor=reduction_time_factor)
if nb_pBLSTM_layers == 1:
print("Encoder has one pBLSTM layers")
print(" hidden_dim", hidden_dim)
if nb_pBLSTM_layers == 2:
print("Encoder has two pBLSTM layers")
print(" hidden_dim", hidden_dim)
reduction_time_factor = pBLSTM_time_reductions[1]
self.pblstm2 = pBLSTM(input_dim=hidden_dim * 2 *
reduction_time_factor,
hidden_dim=hidden_dim,
reduction_time_factor=reduction_time_factor)
print(" hidden_dim", hidden_dim)
elif nb_pBLSTM_layers == 3:
print("Encoder has three pBLSTM layers")
print(" hidden_dim", hidden_dim)
reduction_time_factor = pBLSTM_time_reductions[1]
self.pblstm2 = pBLSTM(input_dim=hidden_dim * 2 *
reduction_time_factor,
hidden_dim=hidden_dim,
reduction_time_factor=reduction_time_factor)
print(" hidden_dim", hidden_dim)
reduction_time_factor = pBLSTM_time_reductions[2]
self.pblstm3 = pBLSTM(input_dim=hidden_dim * 2 *
reduction_time_factor,
hidden_dim=hidden_dim,
reduction_time_factor=reduction_time_factor)
self.key_network = nn.Linear(hidden_dim * 2, value_size)
self.value_network = nn.Linear(hidden_dim * 2, key_size)
