tf.sg_arg_def(frac=(1.0, "test fraction ratio to whole data set. The default is 1.0(=whole set)")) # # hyper parameters # # batch size batch_size = 16 # # inputs # # corpus input tensor ( with QueueRunner ) data = SpeechCorpus(batch_size=batch_size, set_name=tf.sg_arg().set) # mfcc feature of audio x = data.mfcc # target sentence label y = data.label # sequence length except zero-padding seq_len = tf.not_equal(x.sg_sum(axis=2), 0.).sg_int().sg_sum(axis=1) # # Testing Graph # # encode audio feature logit = get_logit(x, voca_size=voca_size)
# set log level to debug
tf.sg_verbosity(10)
#
# hyper parameters
#
batch_size = 16 # total batch size
#
# inputs
#
# corpus input tensor
data = SpeechCorpus(batch_size=batch_size * tf.sg_gpus())
# mfcc feature of audio
inputs = tf.split(data.mfcc, tf.sg_gpus(), axis=0)
# target sentence label
labels = tf.split(data.label, tf.sg_gpus(), axis=0)
# sequence length except zero-padding
seq_len = []
for input_ in inputs:
seq_len.append(
tf.not_equal(input_.sg_sum(axis=2), 0.).sg_int().sg_sum(axis=1))
# parallel loss tower
@tf.sg_parallel