# However, if SGE is used, we cannot simply set CUDA_VISIBLE_DEVICES.
# So it is better to specify the GPU id outside the program.
# Give an arbitrary number (except for -1) to --gpu can enable it. Leave it blank if you want to disable gpu.
import tensorflow as tf
if __name__ == '__main__':
tf.reset_default_graph()
tf.logging.set_verbosity(tf.logging.INFO)
nnet_dir = os.path.join(args.model_dir, "nnet")
config_json = os.path.join(args.model_dir, "nnet/config.json")
if not os.path.isfile(config_json):
sys.exit("Cannot find params.json in %s" % config_json)
params = Params(config_json)
# Change the output node if necessary
if len(args.node) != 0:
params.embedding_node = args.node
tf.logging.info("Extract embedding from %s" % params.embedding_node)
trainer = Trainer(params, args.model_dir, single_cpu=True)
with open(os.path.join(nnet_dir, "feature_dim"), "r") as f:
dim = int(f.readline().strip())
#trainer.build("predict", dim=dim)
trainer.build("predict",
dim=dim,
loss_type="extract_asoftmax",
num_speakers=154)
parser.add_argument("weights", type=str, help="The output weights")
parser.add_argument("embeddings", type=str, help="Embeddings (label vector).")
parser.add_argument("embedding_pic", type=str, help="The output pic")
args = parser.parse_args()
import tensorflow as tf
if __name__ == '__main__':
tf.reset_default_graph()
tf.logging.set_verbosity(tf.logging.INFO)
nnet_dir = os.path.join(args.model_dir, "nnet")
config_json = os.path.join(args.model_dir, "nnet/config.json")
if not os.path.isfile(config_json):
sys.exit("Cannot find params.json in %s" % config_json)
params = Params(config_json)
# params.loss_func = "generalized_angular_triplet_loss"
# params.dict["triplet_center"] = "average"
# params.dict["triplet_center_momentum"] = 0.9
# params.dict["loss_compute"] = "softplus"
# params.dict["margin"] = 0.1
num_total_train_speakers = KaldiDataRandomQueue(
args.data_dir, args.data_spklist).num_total_speakers
dim = FeatureReader(args.data_dir).get_dim()
trainer = Trainer(params, args.model_dir, single_cpu=True)
trainer.build("valid",
dim=dim,
loss_type=params.loss_func,
num_speakers=num_total_train_speakers)
args = parser.parse_args()
if args.gpu == -1:
# Disable GPU
os.environ["CUDA_VISIBLE_DEVICES"] = "-1"
import tensorflow as tf
if __name__ == '__main__':
tf.reset_default_graph()
tf.logging.set_verbosity(tf.logging.INFO)
nnet_dir = os.path.join(args.model_dir, "nnet")
config_json = os.path.join(args.model_dir, "nnet/config.json")
if not os.path.isfile(config_json):
sys.exit("Cannot find params.json in %s" % config_json)
params = Params(config_json)
# Attention weights
params.embedding_node = "attention_weights"
with open(os.path.join(nnet_dir, "feature_dim"), "r") as f:
dim = int(f.readline().strip())
trainer = Trainer(params, args.model_dir, dim, single_cpu=True)
trainer.build("predict")
if args.rspecifier.rsplit(".", 1)[1] == "scp":
# The rspecifier cannot be scp
sys.exit("The rspecifier must be ark or input pipe")
fp_out = open_or_fd(args.wspecifier, "wb")
for index, (key, feature) in enumerate(read_mat_ark(args.rspecifier)):
return mat
if __name__ == '__main__':
reader = tf.train.NewCheckpointReader(nnet_path + 'model-570000')
u, s, v = np.linalg.svd(reader.get_tensor(nsplitname + 'kernel'))
u, s, v = abandon(u, s, v, dimension=0.8)
u = np.mat(u)
s = np.mat(np.diag(np.array(s).squeeze()))
v = np.mat(v)
A = u * s
B = v
C = reader.get_tensor(nsplitname + 'bias')
params = Params(json_path)
x = tf.placeholder(tf.float32, [10, 175, 30], name='x')
features, endpoints = tdnn_svd6(features=x, params=params, mid_channels=A.shape[1])
init = tf.global_variables_initializer()
graph = tf.get_default_graph()
saver = tf.train.Saver()
with tf.Session() as sess:
sess.run(init)
for name in reader.get_variable_to_shape_map():
if nsplitname in name or 'softmax' in name:
continue
herename = nname + re.match(r'(.+?)(\/.*)', name).group(2) + ':0'
sess.run(tf.assign(graph.get_tensor_by_name(herename), reader.get_tensor(name)))
sess.run(tf.assign(graph.get_tensor_by_name(nname + '/tdnn6.0_dense/kernel:0'), A))
sess.run(tf.assign(graph.get_tensor_by_name(nname + '/tdnn6.5_dense/kernel:0'), B))
args = parser.parse_args()
if args.gpu == -1:
# Disable GPU
os.environ["CUDA_VISIBLE_DEVICES"] = "-1"
import tensorflow as tf
if __name__ == '__main__':
tf.reset_default_graph()
tf.logging.set_verbosity(tf.logging.INFO)
nnet_dir = os.path.join(args.model_dir, "nnet")
config_json = os.path.join(args.model_dir, "nnet/config.json")
if not os.path.isfile(config_json):
sys.exit("Cannot find params.json in %s" % config_json)
params = Params(config_json)
# First, we need to extract the weights
num_total_train_speakers = KaldiDataRandomQueue(
os.path.dirname(args.spklist), args.spklist).num_total_speakers
dim = FeatureReader(os.path.dirname(args.spklist)).get_dim()
if "selected_dim" in params.dict:
dim = params.selected_dim
trainer = Trainer(params,
args.model_dir,
dim,
num_total_train_speakers,
single_cpu=True)
trainer.build("valid")
trainer.sess.run(tf.global_variables_initializer())
trainer.sess.run(tf.local_variables_initializer())