import os
import numpy as np
import matplotlib.pyplot as plt
import soundfile as sf
from scipy.io import wavfile
def load_path2gt(paths_file, config):
""" Given the path, construct the ground truth vectors.
This function heavily relies on path2gt_datasets(.),
where the relation between the path and ground truth
are defined.
"""
paths = list()
path2gt = dict()
path2onehot = dict() # REMOVE IF NOT USED!
pf = open(paths_file)
for path in pf.readlines():
path = path.rstrip('\n')
paths.append(path)
label = path2gt_datasets(path, config['dataset'])
path2gt[path] = label
path2onehot[path] = label2onehot(label, config['num_classes_dataset'])
return paths, path2gt, path2onehot
def label2onehot(label, num_classes):
""" Convert class label to one hot vector.
Example: label2onehot(label=2, num_classes=5) > array([0., 0., 1., 0., 0.])
"""
onehot = np.zeros(num_classes)
onehot[label] = 1
return onehot
def path2gt_datasets(path, dataset):
""" Given the audio path, it returns the ground truth label.
Define HERE a new dataset to employ this code with other data.
"""
if dataset == 'GTZAN':
if 'blues' in path:
return 0
elif 'classical' in path:
return 1
elif 'country' in path:
return 2
elif 'disco' in path:
return 3
elif 'hiphop' in path:
return 4
elif 'jazz' in path:
return 5
elif 'metal' in path:
return 6
elif 'pop' in path:
return 7
elif 'reggae' in path:
return 8
elif 'rock' in path:
return 9
else:
print('Did not find the corresponding ground truth (' + str(path) + ')!')
else:
print('Did not find the implementation of ' + str(dataset) + ' dataset!')
def matrix_visualization(matrix,title=None):
""" Visualize 2D matrices like spectrograms or feature maps.
"""
plt.figure()
plt.imshow(np.flipud(matrix.T),interpolation=None)
plt.colorbar()
if title!=None:
plt.title(title)
plt.show()
def wavefile_to_waveform(wav_file, features_type):
data, sr = sf.read(wav_file)
if features_type == 'vggish':
tmp_name = str(int(np.random.rand(1)*1000000)) + '.wav'
sf.write(tmp_name, data, sr, subtype='PCM_16')
sr, wav_data = wavfile.read(tmp_name)
os.remove(tmp_name)
# sr, wav_data = wavfile.read(wav_file) # as done in VGGish Audioset
assert wav_data.dtype == np.int16, 'Bad sample type: %r' % wav_data.dtype
data = wav_data / 32768.0 # Convert to [-1.0, +1.0]
# at least one second of samples, if not repead-pad
src_repeat = data
while (src_repeat.shape[0] < sr):
src_repeat = np.concatenate((src_repeat, data), axis=0)
data = src_repeat[:sr]
return data, sr
