def audio_descriptor(obj, i, a_dictionary, a_noise):
"""
Computes audio descriptor of ith sample of object obj.
:param obj: string. Object from which descriptor is computed.
:param i: int. ith sample of object from which descriptor is computed
"""
filename = globals.path + 'wav/' + obj + '-' + str(i) + '.wav'
#Reads signal
(fs, signal) = au.read(filename)
#Clips signal
decay = range(10, 110, 10)
(RTN, signal, I) = au.RTN(signal, fs, 0.01, decay)
signal = signal[:I[decay.index(60)]]
#Adds additive noise to signal, if a_noise != float('inf'), where a_noise is SNR between signal and noise
signal = au.get_noisy_signal(signal, a_noise, noise_type='white')
signal = au.amplify(signal)
#Gets RTN descriptor from updated signal
(RTN, signal, I) = au.rtn(signal, fs, 0.01, decay)
#audio descriptor
a_descriptor = []
#RTN
a_descriptor.extend(RTN)
#Calculates MFCC descriptor from signal
mfcc = mfcc_descriptor(signal, fs)
b = bof(mfcc, a_dictionary)
a_descriptor.extend(b)
#Calculates FFT descriptor from signal
fftcoefs = au.cft(signal, fs, 0.01, 0.001)
a_descriptor.extend(fftcoefs)
return a_descriptor
src_path = '/Users/avin/git/vc/datasets/timit/TIMIT/TRAIN/*/*'
# src_path = '/Users/avin/git/vc/datasets/kate/sense_and_sensibility_split'
# src_path = '/Users/avin/git/vc/datasets/arctic/bdl'
# src_path = '/Users/avin/git/vc/datasets/kate/therese_raquin_split'
sr = 16000
n_fft = 512
win_length = 400
hop_length = 80
n_sample = 200
amps = []
log_amps = []
dbs = []
for filepath in glob.glob('{}/*.wav'.format(src_path))[:n_sample]:
wav = read(filepath, sr, mono=True)
spec = librosa.stft(wav,
n_fft=n_fft,
win_length=win_length,
hop_length=hop_length) # (n_fft/2+1, t)
amp = np.abs(spec)
amps.extend(amp.flatten())
log_amp = np.log(amp)
log_amps.extend(log_amp.flatten())
db = librosa.amplitude_to_db(amp)
dbs.extend(db.flatten())
amps = np.array(amps)
log_amps = np.array(log_amps)
import librosa.display
import utils
import numpy as np
import matplotlib.pyplot as plt
from audio_utils import read, write
filename = '/Users/avin/git/vc/datasets/timit/TIMIT/TEST/DR1/FAKS0/SA1.wav'
sr = 22050
n_fft = 4096
len_hop = n_fft / 4
plot_wav = True
plot_spec = True
# Waveforms
wav = read(filename, sr, mono=True)
# wav = np.where(wav == 0, 1000, wav)
# wav = np.zeros_like(wav)
# wav[0] = np.ones_like(wav[0])
# Spectrogram
spec = librosa.stft(wav, n_fft=n_fft, hop_length=len_hop)
# Plot waveforms
if plot_wav:
plt.figure(1)
librosa.display.waveplot(wav, sr=sr, color='b')
plt.title('waveform')
plt.tight_layout()
import librosa.display
import utils
import numpy as np
import matplotlib.pyplot as plt
from audio_utils import read, write
filename = '/Users/avin/git/vc/datasets/timit/TIMIT/TEST/DR1/FAKS0/SA1.wav'
sr = 22050
n_fft = 4096
len_hop = n_fft / 4
plot_wav = True
plot_spec = True
# Waveforms
wav = read(filename, sr, mono=True)
# wav = np.where(wav == 0, 1000, wav)
# wav = np.zeros_like(wav)
# wav[0] = np.ones_like(wav[0])
# Spectrogram
spec = librosa.stft(wav, n_fft=n_fft, hop_length=len_hop)
# Plot waveforms
if plot_wav:
plt.figure(1)
librosa.display.waveplot(wav, sr=sr, color='b')
plt.title('waveform')
plt.tight_layout()
def get_descriptors(obj, i, a_dictionary, a_k, a_noise, v_noise, combination):
#Audio filename
filename = globals.path + 'wav/' + obj + '-' + str(i) + '.wav'
#Reads signal
(fs, signal) = au.read(filename)
#Clips signal
decay = range(10, 110, 10)
(RTN, signal, I) = au.RTN(signal, fs, 0.01, decay)
signal = signal[:I[decay.index(60)]]
signal = au.amplify(signal)
audio_descriptor = []
combination = '{:06b}'.format(combination)
#TN
if (int(combination[0])):
audio_descriptor.extend(RTN)
#Stacked MFCC
if (int(combination[1])):
mfcc = mfcc_descriptor(signal, fs)
mfcc = np.reshape(mfcc, -1, order='C')
mfcc = mfcc.tolist()
m.append(len(mfcc))
#mfcc.extend((2938 - len(mfcc))*[0])
audio_descriptor.extend(mfcc[:4056])
#BOF
if (int(combination[2])):
mfcc = mfcc_descriptor(signal, fs)
#mfcc = np.reshape(mfcc, -1, order='C')
#mfcc = mfcc.tolist()
b = bof(mfcc, a_dictionary)
audio_descriptor.extend(b)
#wavelet
if (int(combination[3])):
wavelet = dwt_coefs(signal, a_noise, 'haar', 6)
w.append(len(wavelet))
#wavelet.extend((5050 - len(wavelet))*[0])
audio_descriptor.extend(wavelet[:5050])
#fftIza
if (int(combination[4])):
coefs = au.fft_iza(signal, fs, 0.01, 0.001)
# ciza.append(len(coefs))
# coefs.extend((18225 - len(coefs))*[0])
audio_descriptor.extend(coefs[:25000])
#fftArt
if (int(combination[5])):
coefs = au.cft(signal, fs, 0.01, 0.001)
cart.append(len(coefs))
audio_descriptor.extend(coefs)
#Video
depth = cv.LoadImage(
globals.path + 'img/' + obj + '-' + str(i) + '-depth.png',
cv.CV_LOAD_IMAGE_GRAYSCALE)
mask = cv.LoadImage(globals.path + 'mask.png', cv.CV_LOAD_IMAGE_GRAYSCALE)
if v_noise > 0:
rgb = cv2.imread(
globals.path + 'img/' + obj + '-' + str(i) + '-rgb.png',
cv.CV_LOAD_IMAGE_GRAYSCALE)
noise = np.random.randn(*rgb.shape) * v_noise
# Add this noise to image
noisy = rgb + noise
cv2.imwrite(globals.path + 'noisy.png', noisy)
rgb = cv.LoadImage(globals.path + 'noisy.png',
cv.CV_LOAD_IMAGE_GRAYSCALE)
#rgb = noisy
else:
rgb = cv.LoadImage(
globals.path + 'img/' + obj + '-' + str(i) + '-rgb.png',
cv.CV_LOAD_IMAGE_GRAYSCALE)
#cv.fromarray(rgb)
bridge = CvBridge()
rgb_image = bridge.cv_to_imgmsg(rgb)
depth_image = bridge.cv_to_imgmsg(depth)
mask_image = bridge.cv_to_imgmsg(mask)
rospy.wait_for_service('masked_base_descriptor_service')
try:
descriptor_request = rospy.ServiceProxy(
'masked_base_descriptor_service', masked_service)
response = descriptor_request(rgb=rgb_image,
depth=depth_image,
mask=mask_image)
base_descriptor = bridge.imgmsg_to_cv(response.descriptor)
visual_descriptor = np.array(base_descriptor)
except rospy.ServiceException, e:
print "Service call failed: %s" % e
src_path = '/Users/avin/git/vc/datasets/timit/TIMIT/TRAIN/*/*'
# src_path = '/Users/avin/git/vc/datasets/kate/sense_and_sensibility_split'
# src_path = '/Users/avin/git/vc/datasets/arctic/bdl'
# src_path = '/Users/avin/git/vc/datasets/kate/therese_raquin_split'
sr = 16000
n_fft = 512
win_length = 400
hop_length = 80
n_sample = 200
amps = []
log_amps = []
dbs = []
for filepath in glob.glob('{}/*.wav'.format(src_path))[:n_sample]:
wav = read(filepath, sr, mono=True)
spec = librosa.stft(wav, n_fft=n_fft, win_length=win_length, hop_length=hop_length) # (n_fft/2+1, t)
amp = np.abs(spec)
amps.extend(amp.flatten())
log_amp = np.log(amp)
log_amps.extend(log_amp.flatten())
db = librosa.amplitude_to_db(amp)
dbs.extend(db.flatten())
amps = np.array(amps)
log_amps = np.array(log_amps)
dbs = np.array(dbs)