def testGetDistribution3D(self):
R = 0.0375
H = 0.07
x = np.array([[0, 0, H],
[R, 0, 0],
[R*math.cos(math.pi/3), R*math.sin(math.pi/3), 0],
[-R*math.cos(math.pi/3), R*math.sin(math.pi/3), 0],
[-R, 0, 0],
[-R*math.cos(math.pi/3), -R*math.sin(math.pi/3), 0],
[R*math.cos(math.pi/3), -R*math.sin(math.pi/3), 0]])
nmics = 7
# Setup plot
fig = plt.figure()
ax = fig.add_subplot(111, projection='3d')
plt.show(block=False)
# Peform simulation
N_trials = 10
for n in range(N_trials):
# Get random direction
source = np.array([-100, -100, 0]) + 200 * np.random.rand(3)
#source =np.array([70, 20, 100])
# Compute distances and delays
d = np.sqrt(np.sum((x - source) ** 2, axis=1))
#d1 = d[0] - d
delays = d / pa_tools.SPEED_OF_SOUND
print "delays: " + str(delays)
# Create audio sample
Fs = 44100
T = 1. / Fs
nsecs = .25
N = Fs * nsecs
fund_freq = 50
low_freq = 1 / (2 * nsecs)
n = (np.tile(np.arange(N) * T, (nmics, 1)).T - delays).T
s = np.sin(n * math.pi * low_freq)
# Add different harmonics to signal
for k in range(50):
if k % 3 == 1:
s += 5 * np.sin(n * 2 * math.pi * fund_freq * k)
# Add random noise to each signal
#s += .35 * np.random.rand(nmics, s.shape[1])
# Setup localizer
window_len = 512
N_THETA = 20
N_PHI = N_THETA / 2
loc = DistributionLocalizer(x, sample_rate=Fs, n_theta=N_THETA, dft_len=window_len, n_phi=N_PHI)
# Get section of signal
ind = round(random.random() * (N - 512 - 1))
#ind = 200;
g = s[:, ind:ind + window_len]
#print g
#f = plt.figure()
#a = f.add_subplot(111)
#a.plot(np.arange(g.shape[1]), g.T)
#plt.show()
G = np.fft.fft(g, n=window_len, axis=1)
G_real = np.fft.rfft(g, n=window_len, axis=1)
direcs = loc.get_directions()
d_real = loc.get_distribution_real(G_real)
d = loc.get_distribution_mat(G)
#self.assertListFloatEqual(d, d_real)
d = d_real
print "max: " + str(np.max(d))
print "min: " + str(np.min(d))
maxind = np.argmax(d)
u = 1.5 * direcs[:, maxind]
v = 1.5 * source / np.linalg.norm(source, 2)
#self.assertLessEqual(np.sqrt(np.sum((u / 1.5 - v / 1.5) ** 2)), .2)
plt.cla()
ax.scatter(direcs[0, :], direcs[1, :], direcs[2, :], s=30, c=d)
ax.plot([0, v[0]], [0, v[1]], [0, v[2]])
ax.plot([0, u[0]], [0, u[1]], [0, u[2]], c='r')
#ax.view_init(azim=-90, elev=90)
plt.draw()
time.sleep(.5)
def testGetDistribution3D(self):
R = 0.0375
H = 0.07
x = np.array(
[[0, 0, H], [R, 0, 0],
[R * math.cos(math.pi / 3), R * math.sin(math.pi / 3), 0],
[-R * math.cos(math.pi / 3), R * math.sin(math.pi / 3), 0],
[-R, 0, 0],
[-R * math.cos(math.pi / 3), -R * math.sin(math.pi / 3), 0],
[R * math.cos(math.pi / 3), -R * math.sin(math.pi / 3), 0]])
nmics = 7
# Setup plot
fig = plt.figure()
ax = fig.add_subplot(111, projection='3d')
plt.show(block=False)
# Peform simulation
N_trials = 10
for n in range(N_trials):
# Get random direction
source = np.array([-100, -100, 0]) + 200 * np.random.rand(3)
#source =np.array([70, 20, 100])
# Compute distances and delays
d = np.sqrt(np.sum((x - source)**2, axis=1))
#d1 = d[0] - d
delays = d / pa_tools.SPEED_OF_SOUND
print "delays: " + str(delays)
# Create audio sample
Fs = 44100
T = 1. / Fs
nsecs = .25
N = Fs * nsecs
fund_freq = 50
low_freq = 1 / (2 * nsecs)
n = (np.tile(np.arange(N) * T, (nmics, 1)).T - delays).T
s = np.sin(n * math.pi * low_freq)
# Add different harmonics to signal
for k in range(50):
if k % 3 == 1:
s += 5 * np.sin(n * 2 * math.pi * fund_freq * k)
# Add random noise to each signal
#s += .35 * np.random.rand(nmics, s.shape[1])
# Setup localizer
window_len = 512
N_THETA = 20
N_PHI = N_THETA / 2
loc = DistributionLocalizer(x,
sample_rate=Fs,
n_theta=N_THETA,
dft_len=window_len,
n_phi=N_PHI)
# Get section of signal
ind = round(random.random() * (N - 512 - 1))
#ind = 200;
g = s[:, ind:ind + window_len]
#print g
#f = plt.figure()
#a = f.add_subplot(111)
#a.plot(np.arange(g.shape[1]), g.T)
#plt.show()
G = np.fft.fft(g, n=window_len, axis=1)
G_real = np.fft.rfft(g, n=window_len, axis=1)
direcs = loc.get_directions()
d_real = loc.get_distribution_real(G_real)
d = loc.get_distribution_mat(G)
#self.assertListFloatEqual(d, d_real)
d = d_real
print "max: " + str(np.max(d))
print "min: " + str(np.min(d))
maxind = np.argmax(d)
u = 1.5 * direcs[:, maxind]
v = 1.5 * source / np.linalg.norm(source, 2)
#self.assertLessEqual(np.sqrt(np.sum((u / 1.5 - v / 1.5) ** 2)), .2)
plt.cla()
ax.scatter(direcs[0, :], direcs[1, :], direcs[2, :], s=30, c=d)
ax.plot([0, v[0]], [0, v[1]], [0, v[2]])
ax.plot([0, u[0]], [0, u[1]], [0, u[2]], c='r')
#ax.view_init(azim=-90, elev=90)
plt.draw()
time.sleep(.5)
