def _create_audio_reader(self):
# TODO Calculate receptive_field:
receptive_field = WaveNetModel.calculate_receptive_field(
self.wavenet_params["filter_width"],
self.wavenet_params["dilations"],
self.wavenet_params["scalar_input"],
self.wavenet_params["initial_filter_width"])
# receptive_field = 1
return AudioReader(self.args.audio_dir,
self.coord,
self.args.sample_rate,
self.args.gc_enabled,
receptive_field,
sample_size=self.args.sample_size,
silence_threshold=self.args.silence_threshold,
queue_size=32)
with tf.name_scope('create_inputs'):
# Allow silence trimming to be skipped by specifying a threshold near
# zero.
silence_threshold = None
#AUDIO_FILE_PATH = '/home/sriramso/data/VCTK-Corpus'
AUDIO_FILE_PATH = '/home/andrewszot/VCTK-Corpus'
gc_enabled = False
reader = AudioReader(
AUDIO_FILE_PATH,
coord,
sample_rate=wavenet_params['sample_rate'],
gc_enabled=gc_enabled,
receptive_field=WaveNetModel.calculate_receptive_field(
wavenet_params["filter_width"], wavenet_params["dilations"],
wavenet_params["scalar_input"],
wavenet_params["initial_filter_width"]),
sample_size=39939,
silence_threshold=silence_threshold)
audio_batch = reader.dequeue(1)
if gc_enabled:
gc_id_batch = reader.dequeue_gc(1)
else:
gc_id_batch = None
global_step = tf.Variable(0, trainable=False)
sess = tf.Session(config=tf.ConfigProto(log_device_placement=False))
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
reader.start_threads(sess)