def __init__(self, pooling_type, num_speakers):
super(Tdnn, self).__init__()
if pooling_type == "statistics_pooling":
self.pooling_layer = statistics_pooling()
self.num_speakers = num_speakers
self.layer1 = nn.Sequential(
nn.Conv2d(in_channels=30, out_channels=512, kernel_size=(5, 1)),
nn.BatchNorm2d(512), nn.ReLU())
self.layer2 = nn.Sequential(
nn.Conv2d(in_channels=512, out_channels=512, kernel_size=(5, 1)),
nn.BatchNorm2d(512), nn.ReLU())
self.layer3 = nn.Sequential(
nn.Conv2d(in_channels=512, out_channels=512, kernel_size=(7, 1)),
nn.BatchNorm2d(512), nn.ReLU())
self.layer4 = nn.Sequential(
nn.Linear(in_features=512, out_features=512), nn.BatchNorm1d(512),
nn.ReLU())
self.layer5 = nn.Sequential(
nn.Linear(in_features=512, out_features=1500),
nn.BatchNorm1d(1500), nn.ReLU())
self.layer6 = nn.Linear(in_features=3000, out_features=512)
self.layer6_bn = nn.BatchNorm1d(512)
self.layer6_act = nn.ReLU()
self.layer7 = nn.Sequential(
nn.Linear(in_features=512, out_features=512), nn.BatchNorm1d(512),
nn.ReLU())
self.softmax_layer = nn.Linear(in_features=512,
out_features=num_speakers)
def tdnn_svd(features,
params,
is_training=None,
reuse_variables=None,
aux_features=None):
"""Build a TDNN network.
The structure is similar to Kaldi, while it uses bn+relu rather than relu+bn.
And there is no dilation used, so it has more parameters than Kaldi x-vector.
Args:
features: A tensor with shape [batch, length, dim].
params: Configuration loaded from a JSON.
svd_params: Configuration to point out which layers to be svd loaded from a JSON.
And it should be updated by function "update_mid_channels" before passing to "svdtdnn".
is_training: True if the network is used for training.
reuse_variables: True if the network has been built and enable variable reuse.
aux_features: Auxiliary features (e.g. linguistic features or bottleneck features).
:return:
features: The output of the last layer.
endpoints: An OrderedDict containing output of every components. The outputs are in the order that they add to
the network. Thus it is convenient to split the network by a output name
"""
name = 'tdnn_svd'
svd_json_path = '/data2/liry/test/tf-kaldi-speaker/model/hello.json'
svd_params = Params(svd_json_path)
assert svd_params.updated
# ReLU is a normal choice while other activation function is possible.
relu = tf.nn.relu
for layer in svd_params.split:
if svd_params.split[layer] and svd_params.mid_channels[layer] == -1:
raise AttributeError(
'Please update the mid_channels of %s before construct the graph'
% layer)
if "network_relu_type" in params.dict:
if params.network_relu_type == "prelu":
relu = prelu
if params.network_relu_type == "lrelu":
relu = tf.nn.leaky_relu
endpoints = OrderedDict()
with tf.variable_scope(name, reuse=reuse_variables):
# Convert to [b, 1, l, d]
features = tf.expand_dims(features, 1)
# Layer 1: [-2,-1,0,1,2] --> [b, 1, l-4, 512]
# conv2d + batchnorm + relu
if svd_params.split['tdnn1_conv']:
features = tf.layers.conv2d(
features,
svd_params.mid_channels['tdnn1_conv'], (1, 5),
activation=None,
kernel_regularizer=tf.contrib.layers.l2_regularizer(
params.weight_l2_regularizer),
name='tdnn1.0_conv',
bias_initializer=tf.zeros_initializer())
endpoints["tdnn1.0_conv"] = features
features = tf.layers.conv2d(
features,
32, (1, 1),
activation=None,
kernel_regularizer=tf.contrib.layers.l2_regularizer(
params.weight_l2_regularizer),
name='tdnn1.5_conv')
endpoints["tdnn1.5_conv"] = features
else:
features = tf.layers.conv2d(
features,
32, (1, 5),
activation=None,
kernel_regularizer=tf.contrib.layers.l2_regularizer(
params.weight_l2_regularizer),
name='tdnn1_conv')
endpoints["tdnn1_conv"] = features
features = tf.layers.batch_normalization(
features,
momentum=params.batchnorm_momentum,
training=is_training,
name="tdnn1_bn")
endpoints["tdnn1_bn"] = features
features = relu(features, name='tdnn1_relu')
endpoints["tdnn1_relu"] = features
# Layer 2: [-2, -1, 0, 1, 2] --> [b ,1, l-4, 512]
# conv2d + batchnorm + relu
# This is slightly different with Kaldi which use dilation convolution
if svd_params.split['tdnn2_conv']:
features = tf.layers.conv2d(
features,
svd_params.mid_channels['tdnn2_conv'], (1, 5),
activation=None,
kernel_regularizer=tf.contrib.layers.l2_regularizer(
params.weight_l2_regularizer),
name='tdnn2.0_conv',
bias_initializer=tf.zeros_initializer())
endpoints["tdnn2.0_conv"] = features
features = tf.layers.conv2d(
features,
32, (1, 1),
activation=None,
kernel_regularizer=tf.contrib.layers.l2_regularizer(
params.weight_l2_regularizer),
name='tdnn2.5_conv')
endpoints["tdnn2.5_conv"] = features
else:
features = tf.layers.conv2d(
features,
32, (1, 5),
activation=None,
kernel_regularizer=tf.contrib.layers.l2_regularizer(
params.weight_l2_regularizer),
name='tdnn2_conv')
endpoints["tdnn2_conv"] = features
features = tf.layers.batch_normalization(
features,
momentum=params.batchnorm_momentum,
training=is_training,
name="tdnn2_bn")
endpoints["tdnn2_bn"] = features
features = relu(features, name='tdnn2_relu')
endpoints["tdnn2_relu"] = features
# Layer 3: [-3, -2, -1, 0, 1, 2, 3] --> [b, 1, l-6, 512]
# conv2d + batchnorm + relu
# Still, use a non-dilation one
if svd_params.split['tdnn3_conv']:
features = tf.layers.conv2d(
features,
svd_params.mid_channels['tdnn3_conv'], (1, 7),
activation=None,
kernel_regularizer=tf.contrib.layers.l2_regularizer(
params.weight_l2_regularizer),
name='tdnn3.0_conv',
bias_initializer=tf.zeros_initializer())
endpoints["tdnn3.0_conv"] = features
features = tf.layers.conv2d(
features,
32, (1, 1),
activation=None,
kernel_regularizer=tf.contrib.layers.l2_regularizer(
params.weight_l2_regularizer),
name='tdnn3.5_conv')
endpoints["tdnn3.5_conv"] = features
else:
features = tf.layers.conv2d(
features,
32, (1, 7),
activation=None,
kernel_regularizer=tf.contrib.layers.l2_regularizer(
params.weight_l2_regularizer),
name='tdnn3_conv')
endpoints["tdnn3_conv"] = features
features = tf.layers.batch_normalization(
features,
momentum=params.batchnorm_momentum,
training=is_training,
name="tdnn3_bn")
endpoints["tdnn3_bn"] = features
features = relu(features, name='tdnn3_relu')
endpoints["tdnn3_relu"] = features
# Convert to [b, l, 512]
features = tf.squeeze(features, axis=1)
# The output of the 3-rd layer can simply be rank 3.
endpoints["tdnn3_relu"] = features
# Layer 4: [b, l, 512] --> [b, l, 512]
if svd_params.split['tdnn4_dense']:
features = tf.layers.dense(
features,
svd_params.mid_channels['tdnn4_dense'],
activation=None,
kernel_regularizer=tf.contrib.layers.l2_regularizer(
params.weight_l2_regularizer),
name="tdnn4.0_dense",
bias_initializer=tf.zeros_initializer())
endpoints["tdnn4.0_dense"] = features
features = tf.layers.dense(
features,
512,
activation=None,
kernel_regularizer=tf.contrib.layers.l2_regularizer(
params.weight_l2_regularizer),
name="tdnn4.5_dense")
endpoints["tdnn4.5_dense"] = features
else:
features = tf.layers.dense(
features,
512,
activation=None,
kernel_regularizer=tf.contrib.layers.l2_regularizer(
params.weight_l2_regularizer),
name="tdnn4_dense")
endpoints["tdnn4_dense"] = features
features = tf.layers.batch_normalization(
features,
momentum=params.batchnorm_momentum,
training=is_training,
name="tdnn4_bn")
endpoints["tdnn4_bn"] = features
features = relu(features, name='tdnn4_relu')
endpoints["tdnn4_relu"] = features
# Layer 5: [b, l, x]
if "num_nodes_pooling_layer" not in params.dict:
# The default number of nodes before pooling
params.dict["num_nodes_pooling_layer"] = 1500
if svd_params.split['tdnn5_dense']:
features = tf.layers.dense(
features,
svd_params.mid_channels['tdnn5_dense'],
activation=None,
kernel_regularizer=tf.contrib.layers.l2_regularizer(
params.weight_l2_regularizer),
name="tdnn5.0_dense",
bias_initializer=tf.zeros_initializer())
endpoints["tdnn5.0_dense"] = features
features = tf.layers.dense(
features,
params.num_nodes_pooling_layer,
activation=None,
kernel_regularizer=tf.contrib.layers.l2_regularizer(
params.weight_l2_regularizer),
name="tdnn5.5_dense")
endpoints["tdnn5.5_dense"] = features
else:
features = tf.layers.dense(
features,
params.num_nodes_pooling_layer,
activation=None,
kernel_regularizer=tf.contrib.layers.l2_regularizer(
params.weight_l2_regularizer),
name="tdnn5_dense")
endpoints["tdnn5_dense"] = features
features = tf.layers.batch_normalization(
features,
momentum=params.batchnorm_momentum,
training=is_training,
name="tdnn5_bn")
endpoints["tdnn5_bn"] = features
features = relu(features, name='tdnn5_relu')
endpoints["tdnn5_relu"] = features
# Pooling layer
# If you add new pooling layer, modify this code.
# Statistics pooling
# [b, l, 1500] --> [b, x]
if params.pooling_type == "statistics_pooling":
features = statistics_pooling(features, aux_features, endpoints,
params, is_training)
elif params.pooling_type == "self_attention":
features = self_attention(features, aux_features, endpoints,
params, is_training)
elif params.pooling_type == "ghost_vlad":
features = ghost_vlad(features, aux_features, endpoints, params,
is_training)
# elif params.pooling_type == "aux_attention":
# features = aux_attention(features, aux_features, endpoints, params, is_training)
else:
raise NotImplementedError("Not implement %s pooling" %
params.pooling_type)
endpoints['pooling'] = features
# Utterance-level network
# Layer 6: [b, 512]
if svd_params.split['tdnn6_dense']:
features = tf.layers.dense(
features,
svd_params.mid_channels['tdnn6_dense'],
activation=None,
kernel_regularizer=tf.contrib.layers.l2_regularizer(
params.weight_l2_regularizer),
name='tdnn6.0_dense',
bias_initializer=tf.zeros_initializer())
endpoints['tdnn6.0_dense'] = features
features = tf.layers.dense(
features,
512,
activation=None,
kernel_regularizer=tf.contrib.layers.l2_regularizer(
params.weight_l2_regularizer),
name='tdnn6.5_dense')
endpoints['tdnn6.5_dense'] = features
else:
features = tf.layers.dense(
features,
512,
activation=None,
kernel_regularizer=tf.contrib.layers.l2_regularizer(
params.weight_l2_regularizer),
name='tdnn6_dense')
endpoints['tdnn6_dense'] = features
features = tf.layers.batch_normalization(
features,
momentum=params.batchnorm_momentum,
training=is_training,
name="tdnn6_bn")
endpoints["tdnn6_bn"] = features
features = relu(features, name='tdnn6_relu')
endpoints["tdnn6_relu"] = features
# Layer 7: [b, x]
if "num_nodes_last_layer" not in params.dict:
# The default number of nodes in the last layer
params.dict["num_nodes_last_layer"] = 512
if svd_params.split['tdnn7_dense']:
features = tf.layers.dense(
features,
svd_params.mid_channels['tdnn7_dense'],
activation=None,
kernel_regularizer=tf.contrib.layers.l2_regularizer(
params.weight_l2_regularizer),
name='tdnn7.0_dense',
bias_initializer=tf.zeros_initializer())
endpoints['tdnn7.0_dense'] = features
features = tf.layers.dense(
features,
params.num_nodes_last_layer,
activation=None,
kernel_regularizer=tf.contrib.layers.l2_regularizer(
params.weight_l2_regularizer),
name='tdnn7.5_dense')
endpoints['tdnn7.5_dense'] = features
else:
features = tf.layers.dense(
features,
params.num_nodes_last_layer,
activation=None,
kernel_regularizer=tf.contrib.layers.l2_regularizer(
params.weight_l2_regularizer),
name='tdnn7_dense')
endpoints['tdnn7_dense'] = features
if "last_layer_no_bn" not in params.dict:
params.last_layer_no_bn = False
if not params.last_layer_no_bn:
features = tf.layers.batch_normalization(
features,
momentum=params.batchnorm_momentum,
training=is_training,
name="tdnn7_bn")
endpoints["tdnn7_bn"] = features
if "last_layer_linear" not in params.dict:
params.last_layer_linear = False
if not params.last_layer_linear:
# If the last layer is linear, no further activation is needed.
features = relu(features, name='tdnn7_relu')
endpoints["tdnn7_relu"] = features
return features, endpoints
def tdnn_distill(features, params, is_training=None, reuse_variables=None, aux_features=None):
"""Build a TDNN network.
The structure is similar to Kaldi, while it uses bn+relu rather than relu+bn.
And there is no dilation used, so it has more parameters than Kaldi x-vector.
Args:
features: A tensor with shape [batch, length, dim].
params: Configuration loaded from a JSON.
is_training: True if the network is used for training.
reuse_variables: True if the network has been built and enable variable reuse.
aux_features: Auxiliary features (e.g. linguistic features or bottleneck features).
:return:
features: The output of the last layer.
endpoints: An OrderedDict containing output of every components. The outputs are in the order that they add to
the network. Thus it is convenient to split the network by a output name
"""
# ReLU is a normal choice while other activation function is possible.
relu = tf.nn.relu
if "network_relu_type" in params.dict:
if params.network_relu_type == "prelu":
relu = prelu
if params.network_relu_type == "lrelu":
relu = tf.nn.leaky_relu
endpoints = OrderedDict()
with tf.variable_scope("tdnn", reuse=reuse_variables):
# Convert to [b, 1, l, d]
features = tf.expand_dims(features, 1)
# Layer 1: [-2,-1,0,1,2] --> [b, 1, l-4, 512]
# conv2d + batchnorm + relu
features = tf.layers.conv2d(features,
64,
(1, 5),
activation=None,
kernel_regularizer=tf.contrib.layers.l2_regularizer(params.weight_l2_regularizer),
name='tdnn1_conv')
endpoints["tdnn1_conv"] = features
features = tf.layers.batch_normalization(features,
momentum=params.batchnorm_momentum,
training=is_training,
name="tdnn1_bn")
endpoints["tdnn1_bn"] = features
features = relu(features, name='tdnn1_relu')
endpoints["tdnn1_relu"] = features
# Layer 2: [-2, -1, 0, 1, 2] --> [b ,1, l-4, 512]
# conv2d + batchnorm + relu
# This is slightly different with Kaldi which use dilation convolution
features = tf.layers.conv2d(features,
64,
(1, 5),
activation=None,
kernel_regularizer=tf.contrib.layers.l2_regularizer(params.weight_l2_regularizer),
name='tdnn2_conv')
endpoints["tdnn2_conv"] = features
features = tf.layers.batch_normalization(features,
momentum=params.batchnorm_momentum,
training=is_training,
name="tdnn2_bn")
endpoints["tdnn2_bn"] = features
features = relu(features, name='tdnn2_relu')
endpoints["tdnn2_relu"] = features
# Layer 3: [-3, -2, -1, 0, 1, 2, 3] --> [b, 1, l-6, 512]
# conv2d + batchnorm + relu
# Still, use a non-dilation one
features = tf.layers.conv2d(features,
64,
(1, 7),
activation=None,
kernel_regularizer=tf.contrib.layers.l2_regularizer(params.weight_l2_regularizer),
name='tdnn3_conv')
endpoints["tdnn3_conv"] = features
features = tf.layers.batch_normalization(features,
momentum=params.batchnorm_momentum,
training=is_training,
name="tdnn3_bn")
endpoints["tdnn3_bn"] = features
features = relu(features, name='tdnn3_relu')
endpoints["tdnn3_relu"] = features
# Convert to [b, l, 512]
features = tf.squeeze(features, axis=1)
# The output of the 3-rd layer can simply be rank 3.
endpoints["tdnn3_relu"] = features
# Layer 4: [b, l, 512] --> [b, l, 512]
features = tf.layers.dense(features,
512,
activation=None,
kernel_regularizer=tf.contrib.layers.l2_regularizer(params.weight_l2_regularizer),
name="tdnn4_dense")
endpoints["tdnn4_dense"] = features
features = tf.layers.batch_normalization(features,
momentum=params.batchnorm_momentum,
training=is_training,
name="tdnn4_bn")
endpoints["tdnn4_bn"] = features
features = relu(features, name='tdnn4_relu')
endpoints["tdnn4_relu"] = features
# Layer 5: [b, l, x]
if "num_nodes_pooling_layer" not in params.dict:
# The default number of nodes before pooling
params.dict["num_nodes_pooling_layer"] = 1500
features = tf.layers.dense(features,
params.num_nodes_pooling_layer,
activation=None,
kernel_regularizer=tf.contrib.layers.l2_regularizer(params.weight_l2_regularizer),
name="tdnn5_dense")
endpoints["tdnn5_dense"] = features
features = tf.layers.batch_normalization(features,
momentum=params.batchnorm_momentum,
training=is_training,
name="tdnn5_bn")
endpoints["tdnn5_bn"] = features
features = relu(features, name='tdnn5_relu')
endpoints["tdnn5_relu"] = features
# Pooling layer
# If you add new pooling layer, modify this code.
# Statistics pooling
# [b, l, 1500] --> [b, x]
if params.pooling_type == "statistics_pooling":
features = statistics_pooling(features, aux_features, endpoints, params, is_training)
elif params.pooling_type == "self_attention":
features = self_attention(features, aux_features, endpoints, params, is_training)
elif params.pooling_type == "ghost_vlad":
features = ghost_vlad(features, aux_features, endpoints, params, is_training)
# elif params.pooling_type == "aux_attention":
# features = aux_attention(features, aux_features, endpoints, params, is_training)
else:
raise NotImplementedError("Not implement %s pooling" % params.pooling_type)
endpoints['pooling'] = features
# Utterance-level network
# Layer 6: [b, 512]
features = tf.layers.dense(features,
512,
activation=None,
kernel_regularizer=tf.contrib.layers.l2_regularizer(params.weight_l2_regularizer),
name='tdnn6_dense')
endpoints['tdnn6_dense'] = features
features = tf.layers.batch_normalization(features,
momentum=params.batchnorm_momentum,
training=is_training,
name="tdnn6_bn")
endpoints["tdnn6_bn"] = features
features = relu(features, name='tdnn6_relu')
endpoints["tdnn6_relu"] = features
# Layer 7: [b, x]
if "num_nodes_last_layer" not in params.dict:
# The default number of nodes in the last layer
params.dict["num_nodes_last_layer"] = 64
features = tf.layers.dense(features,
params.num_nodes_last_layer,
activation=None,
kernel_regularizer=tf.contrib.layers.l2_regularizer(params.weight_l2_regularizer),
name='tdnn7_dense')
endpoints['tdnn7_dense'] = features
if "last_layer_no_bn" not in params.dict:
params.last_layer_no_bn = False
if not params.last_layer_no_bn:
features = tf.layers.batch_normalization(features,
momentum=params.batchnorm_momentum,
training=is_training,
name="tdnn7_bn")
endpoints["tdnn7_bn"] = features
if "last_layer_linear" not in params.dict:
params.last_layer_linear = False
if not params.last_layer_linear:
# If the last layer is linear, no further activation is needed.
features = relu(features, name='tdnn7_relu')
endpoints["tdnn7_relu"] = features
return features, endpoints
