def linear2mu(signal, mu=255):
"""
-1 < signal < 1
"""
assert signal.max() <= 1 and signal.min() >= -1
y = np.sign(signal) * np.log(1.0 + mu * np.abs(signal)) / np.log(mu + 1)
return y
def compute(self):
signal = self.get_input("Input").get_array().squeeze()
print "signal.shape = ", signal.shape
lof = self.get_input("Low Freq")
hif = self.get_input("High Freq")
num_scalar_bins = self.get_input("Scalar Bins")
num_pts = signal.size
min_s = signal.min()
max_s = signal.max()
d = max_s - min_s
print "Accumulating over " + str(num_pts) + " points"
histo = numpy.zeros((num_scalar_bins,hif-lof))
f_sig = scipy.fftpack.fftn(signal)
for z in range(signal.shape[0]):
for y in range(signal.shape[1]):
for x in range(signal.shape[2]):
yray = f_sig[z,:,x].squeeze()
ty = self.stockwell(yray, lof, hif)
zray = f_sig[:,y,x].squeeze()
tz = self.stockwell(zray, lof, hif)
xray = f_sig[z,y,:].squeeze()
tx = self.stockwell(xray, lof, hif)
tx = tx[:,z].squeeze()
ty = ty[:,y].squeeze()
tz = tz[:,x].squeeze()
tx = tx * tx.conjugate()
ty = ty * ty.conjugate()
tz = tz * tz.conjugate()
ar = tx.real + ty.real + tz.real
ar = ar / ar.sum()
scalar = signal[z,y,x]
try:
bin = int((scalar - min_s) / d * float(num_scalar_bins-1.))
# dist[bin] += 1
# print bin, scalar, ar
sigma = self.force_get_input("Sigma")
if sigma:
ar = scipy.signal.gaussian(ar.size, sigma) * ar
histo[bin,:] += ar
except:
print "Cannot assign to bin: " + str(bin) +", scalar: " +str(scalar)
print "location = ", x, y, z
raise ModuleError("Cannot assign to bin: " + str(bin) +", scalar: " +str(scalar))
# print "done with y = ", y
print "done with z = ", z
out = NDArray()
out.set_array(histo)
self.set_output("Output", out)
def compute(self):
signal = self.getInputFromPort("Input").get_array()
lof = self.getInputFromPort("Low Freq")
hif = self.getInputFromPort("High Freq")
num_pts = signal.size
min_s = signal.min()
max_s = signal.max()
d = max_s - min_s
print "Accumulating over " + str(num_pts) + " points"
histo = numpy.zeros((512, hif - lof + 1))
print "Histo: ", histo.shape
dist = numpy.zeros(512)
for z in range(signal.shape[2]):
for y in range(signal.shape[1]):
for x in range(signal.shape[0]):
sigx = signal[z, y, :]
sigy = signal[z, :, x]
tx = st.st(sigx, lof, hif)
ty = st.st(sigy, lof, hif)
sigz = signal[:, y, z]
tz = st.st(sigz, lof, hif)
tz = tz[:, x].squeeze()
tz = tz * tz.conjugate()
tx = tx[:, z].squeeze()
ty = ty[:, y].squeeze()
tx = tx * tx.conjugate()
ty = ty * ty.conjugate()
ar = tx.real + ty.real + tz.real
ar = ar / ar.sum()
# ar = ar.sum(axis=1)
# print "ar: ", ar.shape
# ar.shape = (ar.shape[0],1)
scalar = signal[z, y, x]
try:
bin = int((scalar - min_s) / d * 511.)
# dist[bin] += 1
# print bin, scalar, ar
sigma = self.forceGetInputFromPort("Sigma")
if sigma:
ar = scipy.signal.gaussian(ar.size, sigma) * ar
histo[bin, :] += ar
except:
print "Cannot assign to bin: " + str(
bin) + ", scalar: " + str(scalar)
print "location = ", x, y, z
raise ModuleError("Cannot assign to bin: " + str(bin) +
", scalar: " + str(scalar))
# print "done with y = ", y
print "done with z = ", z
out = NDArray()
out.set_array(histo) # / dist)
self.setResult("Output", out)
def compute(self):
signal = self.get_input("Input").get_array()
lof = self.get_input("Low Freq")
hif = self.get_input("High Freq")
num_pts = signal.size
min_s = signal.min()
max_s = signal.max()
d = max_s - min_s
print "Accumulating over " + str(num_pts) + " points"
histo = numpy.zeros((512,hif-lof+1))
print "Histo: ",histo.shape
dist = numpy.zeros(512)
for z in range(signal.shape[2]):
for y in range(signal.shape[1]):
for x in range(signal.shape[0]):
sigx = signal[z,y,:]
sigy = signal[z,:,x]
tx = st.st(sigx, lof, hif)
ty = st.st(sigy, lof, hif)
sigz = signal[:,y,z]
tz = st.st(sigz, lof, hif)
tz = tz[:,x].squeeze()
tz = tz * tz.conjugate()
tx = tx[:,z].squeeze()
ty = ty[:,y].squeeze()
tx = tx * tx.conjugate()
ty = ty * ty.conjugate()
ar = tx.real + ty.real + tz.real
ar = ar / ar.sum()
# ar = ar.sum(axis=1)
# print "ar: ", ar.shape
# ar.shape = (ar.shape[0],1)
scalar = signal[z,y,x]
try:
bin = int((scalar - min_s) / d * 511.)
# dist[bin] += 1
# print bin, scalar, ar
sigma = self.force_get_input("Sigma")
if sigma:
ar = scipy.signal.gaussian(ar.size, sigma) * ar
histo[bin,:] += ar
except:
print "Cannot assign to bin: " + str(bin) +", scalar: " +str(scalar)
print "location = ", x, y, z
raise ModuleError("Cannot assign to bin: " + str(bin) +", scalar: " +str(scalar))
# print "done with y = ", y
print "done with z = ", z
out = NDArray()
out.set_array(histo)# / dist)
self.set_output("Output", out)
def normalize(signal):
"""Normalize signal between 0 and 1
Args:
signal (np.array/th.tensor): Signal to normalize
Returns:
np.array/th.tensor: Normalized signal
"""
signal -= signal.min()
signal /= signal.max()
return signal
def waveform_shaded(signal: np.ndarray, fs: int, start=0, end=None):
if end is None:
end = len(signal) / fs
ymargin_factor = 1.1
x_range = [start, end]
y_range = [ymargin_factor * signal.min(), ymargin_factor * signal.max()]
t = np.linspace(start, end, num=len(signal))
df = pd.DataFrame(data={'Time': t, 'Signal': signal})
cvs = ds.Canvas(x_range=x_range, y_range=y_range, plot_width=1500)
cols = ['Signal']
aggs = OrderedDict((c, cvs.line(df, 'Time', c)) for c in cols)
img = tf.shade(aggs['Signal'])
arr = np.array(img)
z = arr.tolist()
dims = len(z[0]), len(z)
x = np.linspace(x_range[0], x_range[1], dims[0])
y = np.linspace(y_range[0], y_range[1], dims[0])
fig = {
'data': [{
'x': x,
'y': y,
'z': z,
'type': 'heatmap',
'showscale': False,
'colorscale': [[0, 'rgba(255, 255, 255,0)'], [1, '#75baf2']],
}],
'layout': {
'height': 350,
'xaxis': {
'title': 'Time [s]',
'showline': True,
'zeroline': False,
'showgrid': False,
'showticklabels': True
},
'yaxis': {
'title': 'Amplitude',
'fixedrange': True,
'showline': False,
'zeroline': False,
'showgrid': False,
'showticklabels': False,
'ticks': ''
},
'title': 'Waveform'
}
}
return fig
def plot_signal_attr(fig, ax, attr, signal, fs=1.0, filter=True, lw=1.0):
# Plots a signal with explanation strength as sample colors
if filter:
signal = bandpass_filter(signal, fs, fc=[1, 30])
t = np.linspace(0, len(signal) / fs, len(signal))
points = np.array([t, signal]).T.reshape(-1, 1, 2)
segments = np.concatenate([points[:-1], points[1:]], axis=1)
# Create a continuous norm to map from data points to colors
norm = plt.Normalize(attr.min(), attr.max())
lc = LineCollection(segments, cmap='inferno_r', norm=norm)
# Set the values used for colormapping
lc.set_array(attr)
lc.set_linewidth(lw)
line = ax.add_collection(lc)
ax.set_xlim(t.min(), t.max())
ax.set_ylim(signal.min(), signal.max())
ax.get_yaxis().set_ticks([])
ax.get_xaxis().set_ticks([])
def compute(self):
signal = self.getInputFromPort("Input").get_array().squeeze()
print "signal.shape = ", signal.shape
lof = self.getInputFromPort("Low Freq")
hif = self.getInputFromPort("High Freq")
num_scalar_bins = self.getInputFromPort("Scalar Bins")
num_pts = signal.size
min_s = signal.min()
max_s = signal.max()
d = max_s - min_s
print "Accumulating over " + str(num_pts) + " points"
histo = numpy.zeros((num_scalar_bins, hif - lof))
f_sig = scipy.fftpack.fftn(signal)
for z in range(signal.shape[0]):
for y in range(signal.shape[1]):
for x in range(signal.shape[2]):
yray = f_sig[z, :, x].squeeze()
ty = self.stockwell(yray, lof, hif)
zray = f_sig[:, y, x].squeeze()
tz = self.stockwell(zray, lof, hif)
xray = f_sig[z, y, :].squeeze()
tx = self.stockwell(xray, lof, hif)
tx = tx[:, z].squeeze()
ty = ty[:, y].squeeze()
tz = tz[:, x].squeeze()
tx = tx * tx.conjugate()
ty = ty * ty.conjugate()
tz = tz * tz.conjugate()
ar = tx.real + ty.real + tz.real
ar = ar / ar.sum()
scalar = signal[z, y, x]
try:
bin = int(
(scalar - min_s) / d * float(num_scalar_bins - 1.))
# dist[bin] += 1
# print bin, scalar, ar
sigma = self.forceGetInputFromPort("Sigma")
if sigma:
ar = scipy.signal.gaussian(ar.size, sigma) * ar
histo[bin, :] += ar
except:
print "Cannot assign to bin: " + str(
bin) + ", scalar: " + str(scalar)
print "location = ", x, y, z
raise ModuleError("Cannot assign to bin: " + str(bin) +
", scalar: " + str(scalar))
# print "done with y = ", y
print "done with z = ", z
out = NDArray()
out.set_array(histo)
self.setResult("Output", out)
def generate_feat_opts(path=None,
cfg={
'pkg': 'pysp',
'type': 'logfbank',
'nfilt': 40,
'delta': 2
},
signal=None,
rate=16000):
cfg = dict(cfg)
if cfg['pkg'] == 'pysp': # python_speech_features #
if signal is None:
rate, signal = wavfile.read(path)
if cfg['type'] == 'logfbank':
feat_mat = pyspfeat.base.logfbank(signal,
rate,
nfilt=cfg.get('nfilt', 40))
elif cfg['type'] == 'mfcc':
feat_mat = pyspfeat.base.mfcc(signal,
rate,
numcep=cfg.get('nfilt', 26) // 2,
nfilt=cfg.get('nfilt', 26))
elif cfg['type'] == 'wav':
feat_mat = pyspfeat.base.sigproc.framesig(
signal,
frame_len=cfg.get('frame_len', 400),
frame_step=cfg.get('frame_step', 160))
else:
raise NotImplementedError(
"feature type {} is not implemented/available".format(
cfg['type']))
pass
# delta #
comb_feat_mat = [feat_mat]
delta = cfg['delta']
if delta > 0:
delta_feat_mat = pyspfeat.base.delta(feat_mat, 2)
comb_feat_mat.append(delta_feat_mat)
if delta > 1:
delta2_feat_mat = pyspfeat.base.delta(delta_feat_mat, 2)
comb_feat_mat.append(delta2_feat_mat)
if delta > 2:
raise NotImplementedError(
"max delta is 2, larger than 2 is not normal setting")
return np.hstack(comb_feat_mat)
elif cfg['pkg'] == 'rosa':
if signal is None:
signal, rate = librosa.core.load(path, sr=cfg['sample_rate'])
assert rate == cfg[
'sample_rate'], "sample rate is different with current data"
if cfg.get('preemphasis', None) is not None:
# signal = np.append(signal[0], signal[1:] - cfg['preemphasis']*signal[:-1])
signal = signal_util.preemphasis(x, self.cfg['preemphasis'])
if cfg.get('pre', None) == 'meanstd':
signal = (signal - signal.mean()) / signal.std()
elif cfg.get('pre', None) == 'norm':
signal = (signal - signal.min()) / (signal.max() -
signal.min()) * 2 - 1
# raw feature
if cfg['type'] == 'wav':
if cfg.get('post', None) == 'mu':
signal = linear2mu(signal)
feat_mat = pyspfeat.base.sigproc.framesig(
signal,
frame_len=cfg.get('frame_len', 400),
frame_step=cfg.get('frame_step', 160))
return feat_mat
# spectrogram-based feature
raw_spec = signal_util.rosa_spectrogram(
signal,
n_fft=cfg['nfft'],
hop_length=cfg.get('winstep', None),
win_length=cfg.get('winlen', None))[0]
if cfg['type'] in ['logmelfbank', 'melfbank']:
mel_spec = signal_util.rosa_spec2mel(raw_spec, nfilt=cfg['nfilt'])
if cfg['type'] == 'logmelfbank':
return np.log(mel_spec)
else:
return mel_spec
elif cfg['type'] == 'lograwfbank':
return np.log(raw_spec)
elif cfg['type'] == 'rawfbank':
return raw_spec
else:
raise NotImplementedError()
elif cfg['pkg'] == 'taco':
# SPECIAL FOR TACOTRON #
tacohelper = TacotronHelper(cfg)
if signal is None:
signal = tacohelper.load_wav(path)
assert len(signal) != 0, ('file {} is empty'.format(path))
try:
if cfg['type'] == 'raw':
feat = tacohelper.spectrogram(signal).T
elif cfg['type'] == 'mel':
feat = tacohelper.melspectrogram(signal).T
else:
raise NotImplementedError()
except:
import ipdb
ipdb.set_trace()
pass
return feat
elif cfg['pkg'] == 'world':
if path is None:
with tempfile.NamedTemporaryFile() as tmpfile:
wavfile.write(tmpfile.name, rate, signal)
logf0, bap, mgc = world_vocoder_util.world_analysis(
tmpfile.name, cfg['mcep'])
else:
logf0, bap, mgc = world_vocoder_util.world_analysis(
path, cfg['mcep'])
vuv, f0, bap, mgc = world_vocoder_util.world2feat(logf0, bap, mgc)
# ignore delta, avoid curse of dimensionality #
return vuv, f0, bap, mgc
else:
raise NotImplementedError()
pass
