def blow_up(F, x, y, aspect, ax, zoom=None):
w = x[1]+1-x[0]
h = y[1]+1-y[0]
extent = [0,w,0,h]
plt.imshow(u.dB(F[x[0]:x[1]+1,y[0]:y[1]+1].T),
aspect=aspect,
origin='lower', extent=extent,
vmin=vmin, vmax=vmax, cmap=cmap, interpolation=interpolation)
if zoom is not None:
wo = w*(1-zoom)/zoom
ho = h*(1-zoom)/zoom
ax.set_xlim(-wo/2,w+wo/2)
ax.set_ylim(-ho/2,h+ho/2)
ax.set_ylabel('')
ax.set_xlabel('')
ax.axis('off')
def blow_up(F, x, y, aspect, ax, zoom=None):
w = x[1] + 1 - x[0]
h = y[1] + 1 - y[0]
extent = [0, w, 0, h]
plt.imshow(u.dB(F[x[0]:x[1] + 1, y[0]:y[1] + 1].T),
aspect=aspect,
origin='lower',
extent=extent,
vmin=vmin,
vmax=vmax,
cmap=cmap,
interpolation=interpolation)
if zoom is not None:
wo = w * (1 - zoom) / zoom
ho = h * (1 - zoom) / zoom
ax.set_xlim(-wo / 2, w + wo / 2)
ax.set_ylim(-ho / 2, h + ho / 2)
ax.set_ylabel('')
ax.set_xlabel('')
ax.axis('off')
opesq_tri = np.zeros((0,2,2))
opesq_bf = np.zeros((0,2,NBF,max_sources))
isinr = np.zeros((0))
osinr_tri = np.zeros((0,2))
osinr_bf = np.zeros((0,NBF,max_sources))
# Read in all the data
for fname in files:
print 'Loading from',fname
a = np.load(fname)
good_source = np.concatenate((good_source, a['good_source']), axis=0)
bad_source = np.concatenate((bad_source, a['bad_source']), axis=0)
isinr = np.concatenate((isinr,u.dB(a['isinr'])), axis=0)
osinr_bf = np.concatenate((osinr_bf,u.dB(a['osinr_bf'])), axis=0)
osinr_tri = np.concatenate((osinr_tri,u.dB(a['osinr_trinicon'])), axis=0)
ipesq = np.concatenate((ipesq,a['pesq_input']), axis=0)
opesq_bf = np.concatenate((opesq_bf,a['pesq_bf']), axis=0)
opesq_tri = np.concatenate((opesq_tri,a['pesq_trinicon']), axis=0)
loops = good_source.shape[0]
print 'Number of loops:',loops
print 'Median input Raw MOS',np.median(ipesq[:,0])
print 'Median input MOS LQO',np.median(ipesq[:,1])
print 'Median input SINR',np.median(isinr[:])
# Trinicon is blind so we have PESQ for both output channels
# Select the channel that has highest Raw MOS for evaluation
mics.rakeMaxSINRWeights(good_sources, bad_sources,
R_n = sigma2_n*np.eye(mics.M),
rcond=0.,
attn=True, ff=False)
output_maxsinr = mics.process()
# high-pass and normalize
output_maxsinr = u.highpass(output_maxsinr, Fs)
output_maxsinr = u.normalize(output_maxsinr)
'''
PLOT SPECTROGRAM
'''
dSNR = u.dB(room1.dSNR(mics.center[:,0], source=0), power=True)
print 'The direct SNR for good source is ' + str(dSNR)
# as comparison pic central mic signal
input_mic = mics.signals[mics.M/2]
# high-pass and normalize
input_mic = u.highpass(input_mic, Fs)
input_mic = u.normalize(input_mic)
# remove a bit of signal at the end and time-align all signals.
# the delays were visually measured by plotting the signals
n_lim = np.ceil(len(input_mic) - t_cut*Fs)
input_clean = signal1[:n_lim]
input_mic = input_mic[105:n_lim+105]
output_mvdr = output_mvdr[31:n_lim+31]
opesq_tri = np.zeros((0, 2, 2))
opesq_bf = np.zeros((0, 2, NBF, max_sources))
isinr = np.zeros((0))
osinr_tri = np.zeros((0, 2))
osinr_bf = np.zeros((0, NBF, max_sources))
# Read in all the data
for fname in files:
print 'Loading from', fname
a = np.load(fname)
good_source = np.concatenate((good_source, a['good_source']), axis=0)
bad_source = np.concatenate((bad_source, a['bad_source']), axis=0)
isinr = np.concatenate((isinr, u.dB(a['isinr'])), axis=0)
osinr_bf = np.concatenate((osinr_bf, u.dB(a['osinr_bf'])), axis=0)
osinr_tri = np.concatenate((osinr_tri, u.dB(a['osinr_trinicon'])), axis=0)
ipesq = np.concatenate((ipesq, a['pesq_input']), axis=0)
opesq_bf = np.concatenate((opesq_bf, a['pesq_bf']), axis=0)
opesq_tri = np.concatenate((opesq_tri, a['pesq_trinicon']), axis=0)
loops = good_source.shape[0]
print 'Number of loops:', loops
print 'Median input Raw MOS', np.median(ipesq[:, 0])
print 'Median input MOS LQO', np.median(ipesq[:, 1])
print 'Median input SINR', np.median(isinr[:])
# Trinicon is blind so we have PESQ for both output channels
# Select the channel that has highest Raw MOS for evaluation
bad_sources,
R_n=sigma2_n * np.eye(mics.M),
rcond=0.,
attn=True,
ff=False)
output_maxsinr = mics.process()
# high-pass and normalize
output_maxsinr = u.highpass(output_maxsinr, Fs)
output_maxsinr = u.normalize(output_maxsinr)
'''
PLOT SPECTROGRAM
'''
dSNR = u.dB(room1.dSNR(mics.center[:, 0], source=0), power=True)
print 'The direct SNR for good source is ' + str(dSNR)
# as comparison pic central mic signal
input_mic = mics.signals[mics.M / 2]
# high-pass and normalize
input_mic = u.highpass(input_mic, Fs)
input_mic = u.normalize(input_mic)
# remove a bit of signal at the end and time-align all signals.
# the delays were visually measured by plotting the signals
n_lim = np.ceil(len(input_mic) - t_cut * Fs)
input_clean = signal1[:n_lim]
input_mic = input_mic[105:n_lim + 105]
output_mvdr = output_mvdr[31:n_lim + 31]