Coverage for hypergan/inputs/audio_loader.py : 20%

import glob 

import tensorflow as tf 

import hypergan.inputs.resize_audio_patch 

from tensorflow.contrib import ffmpeg 

class AudioLoader: 

def build_labels(dirs): 

next_id=0 

labels = {} 

for dir in dirs: 

labels[dir.split('/')[-1]]=next_id 

next_id+=1 

return labels,next_id 

def mp3_tensors_from_directory(directory, batch_size, channels=2, format='mp3', seconds=30, bitrate=16384): 

filenames = glob.glob(directory+"/**/*."+format) 

labels,total_labels = build_labels(sorted(glob.glob(directory+"/*"))) 

num_examples_per_epoch = 10000 

# Create a queue that produces the filenames to read. 

classes = [labels[f.split('/')[-2]] for f in filenames] 

print("Found files", len(filenames)) 

filenames = tf.convert_to_tensor(filenames, dtype=tf.string) 

classes = tf.convert_to_tensor(classes, dtype=tf.int32) 

print("[0]", filenames[0], classes[0]) 

input_queue = tf.train.slice_input_producer([filenames, classes]) 

# Read examples from files in the filename queue. 

print("INPUT_QUEUE", input_queue[0]) 

value = tf.read_file(input_queue[0]) 

#preprocess = tf.read_file(input_queue[0]+'.preprocess') 

print("Preloaded data", value) 

#print("Loaded data", data) 

label = input_queue[1] 

min_fraction_of_examples_in_queue = 0.4 

min_queue_examples = int(num_examples_per_epoch * 

min_fraction_of_examples_in_queue) 

#data = tf.cast(data, tf.float32) 

data = ffmpeg.decode_audio(value, file_format=format, samples_per_second=bitrate, channel_count=channels) 

data = shared.resize_audio_patch.resize_audio_with_crop_or_pad(data, seconds*bitrate*channels, 0,True) 

#data = tf.slice(data, [0,0], [seconds*bitrate, channels]) 

tf.Tensor.set_shape(data, [seconds*bitrate, channels]) 

#data = tf.minimum(data, 1) 

#data = tf.maximum(data, -1) 

data = data/tf.reduce_max(tf.reshape(tf.abs(data),[-1])) 

print("DATA IS", data) 

x,y=_get_data(data, label, min_queue_examples, batch_size) 

return x, y, total_labels, num_examples_per_epoch 

def _get_data(image, label, min_queue_examples, batch_size): 

num_preprocess_threads = 1 

print(image, label) 

images, label_batch= tf.train.shuffle_batch( 

[image, label], 

batch_size=batch_size, 

num_threads=num_preprocess_threads, 

capacity= 502, 

min_after_dequeue=128) 

return images, tf.reshape(label_batch, [batch_size])