def testGetCamDir2Plane2(self):
planes = [self.plane1, self.plane2]
mic_loc = np.array([0, 0, 0])
cam_loc = np.array([1, 1, 1])
space = SearchSpace(mic_loc, cam_loc, planes)
direction = np.array([-1, 1, 1])
camdir = space.get_camera_dir(direction)
self.assertListEqual(list(camdir), list(np.array([-6, 4, 4]) / (68 ** .5)))
def testGetCameraDir3Plane(self):
planes = [self.plane1, self.plane2, self.plane3]
mic_loc = np.array([0, 0, 0])
cam_loc = np.array([1, 1, 1])
space = SearchSpace(mic_loc, cam_loc, planes)
direction = np.array([-1, 1, 1])
camdir = space.get_camera_dir(direction)
self.assertListEqual(list(camdir), [-1, 0, 0])
def testGetCameraDir3Plane(self):
planes = [self.plane1, self.plane2, self.plane3]
mic_loc = np.array([0, 0, 0])
cam_loc = np.array([1, 1, 1])
space = SearchSpace(mic_loc, cam_loc, planes)
direction = np.array([-1, 1, 1])
camdir = space.get_camera_dir(direction)
self.assertListEqual(list(camdir), [-1, 0, 0])
def testGetCamDir2Plane2(self):
planes = [self.plane1, self.plane2]
mic_loc = np.array([0, 0, 0])
cam_loc = np.array([1, 1, 1])
space = SearchSpace(mic_loc, cam_loc, planes)
direction = np.array([-1, 1, 1])
camdir = space.get_camera_dir(direction)
self.assertListEqual(list(camdir),
list(np.array([-6, 4, 4]) / (68**.5)))
def localize():
# Setup search space
# x vector points to front of class, -z vector points to floor
teacher_plane = SourcePlane(TEACHER_NORMAL, TEACHER_OFFSET)
student_plane = SourcePlane(STUDENT_NORMAL, STUDENT_OFFSET)
space = SearchSpace(MIC_LOC, CAMERA_LOC, [teacher_plane, student_plane])
# Setup camera
forward = np.array([1, 0, 0])
above = np.array([0, 0, 1])
camera = SonyCamera(URL, forward, above)
# Setup pyaudio instances
pa = pyaudio.PyAudio()
helper = AudioHelper(pa)
localizer = DistributionLocalizer(mic_positions=mic_layout,
dft_len=FFT_LENGTH,
sample_rate=SAMPLE_RATE,
n_theta=N_THETA,
n_phi=N_PHI)
# Setup STFT object
stft = StftManager(dft_length=FFT_LENGTH,
window_length=WINDOW_LENGTH,
hop_length=HOP_LENGTH,
use_window_fcn=True,
n_channels=NUM_CHANNELS_IN,
dtype=DATA_TYPE)
# Setup devices
in_device = helper.get_input_device_from_user()
if PLAY_AUDIO:
out_device = helper.get_output_device_from_user()
else:
out_device = helper.get_default_output_device_info()
# Setup streams
in_stream = pa.open(rate=SAMPLE_RATE,
channels=NUM_CHANNELS_IN,
format=SAMPLE_TYPE,
frames_per_buffer=FRAMES_PER_BUF,
input=True,
input_device_index=int(in_device['index']),
stream_callback=read_in_data)
out_stream = pa.open(rate=SAMPLE_RATE,
channels=NUM_CHANNELS_OUT,
format=SAMPLE_TYPE,
output=True,
frames_per_buffer=FRAMES_PER_BUF,
output_device_index=int(out_device['index']),
stream_callback=write_out_data)
# Start recording/playing back
in_stream.start_stream()
out_stream.start_stream()
# Start thread to check for user quit
quit_thread = threading.Thread(target=check_for_quit)
quit_thread.start()
# Plotting setup
if PLOT_CARTES:
fig = plt.figure()
ax = fig.add_subplot(111, projection='3d')
plt.show(block=False)
scat = []
if PLOT_SPACE:
fig = plt.figure()
ax = fig.add_subplot(111, projection='3d')
# Setup bounds
xlo, xhi = (-5, DISTANCE_TO_TEACHER + 5)
ylo, yhi = (-15, 15)
zlo, zhi = (-15, 5)
# Setup grid
nx, ny = (200, 100)
x = np.linspace(xlo, xhi, nx)
y = np.linspace(ylo, yhi, ny)
X, Y = np.meshgrid(x, y)
n, m = (STUDENT_NORMAL, STUDENT_OFFSET)
TP = (n.dot(m) - n[0] * X - n[1] * Y) / n[2] - 2
# Plot markers for mic
m = MIC_LOC
ax.plot([MIC_LOC[0]], [MIC_LOC[1]], [MIC_LOC[2]], 'r.', markersize=10.)
# Plot marker for camera
c = CAMERA_LOC
ax.plot([CAMERA_LOC[0]], [CAMERA_LOC[1]], [CAMERA_LOC[2]], 'b.', markersize=10.)
# Draw lines from camera and mic to source
source_loc = np.array([10, 0, 0])
source_point, = ax.plot([source_loc[0]], [source_loc[1]], [source_loc[2]], 'black', marker='.', markersize=10.)
s = source_loc
camera_dir, = ax.plot([c[0], m[0]], [c[1], m[1]], [c[2], m[2]], 'blue')
mic_dir, = ax.plot([m[0], m[0]], [m[1], m[1]], [m[2], m[2]], 'red')
#ax.plot_surface(X, Y, TP)
ax.set_xlim(xlo, xhi)
ax.set_ylim(ylo, yhi)
ax.set_zlim(zlo, zhi)
ax.view_init(elev=25, azim=-120)
plt.show(block=False)
if EXTERNAL_PLOT:
fig = plt.figure()
ax = fig.add_subplot(111)
plt.show(block=False)
count = 0
prev_direc = np.array([0, 0, 0])
direcs = localizer.get_directions()
try:
global done
while in_stream.is_active() or out_stream.is_active():
data_available = in_buf.wait_for_read(WINDOW_LENGTH, TIMEOUT)
if data_available:
# Get data from the circular buffer
data = in_buf.read_samples(WINDOW_LENGTH)
# Perform an stft
stft.performStft(data)
# Process dfts from windowed segments of input
dfts = stft.getDFTs()
d = localizer.get_3d_real_distribution(dfts)
ind = np.argmax(d)
u = 1.5 * direcs[:, ind] # Direction of arrival
if DO_TRACK and count % TRACKING_FREQ == 0:
#v = np.array([1, 0, 1])
v = u
direc = space.get_camera_dir(v)
if not direc.any():
direc = prev_direc
else:
prev_direc = direc
# Send camera new direction
camera.face_direction(direc)
if PLOT_SPACE:
if direc.any():
src = space.get_source_loc(u)
source_point.set_xdata([src[0]])
source_point.set_ydata([src[1]])
source_point.set_3d_properties(zs=[src[2]])
cam_src = CAMERA_LOC + 30 * direc
mic_src = MIC_LOC + 30 * u
# Update camera line
camera_dir.set_xdata([CAMERA_LOC[0], cam_src[0]])
camera_dir.set_ydata([CAMERA_LOC[1], cam_src[1]])
camera_dir.set_3d_properties(zs=[CAMERA_LOC[2], cam_src[2]])
# Update mic line
mic_dir.set_xdata([MIC_LOC[0], mic_src[0]])
mic_dir.set_ydata([MIC_LOC[1], mic_src[1]])
mic_dir.set_3d_properties(zs=[MIC_LOC[2], mic_src[2]])
plt.draw()
# Take care of plotting
if count % 1 == 0:
if PLOT_CARTES:
plt.cla()
ax.scatter(direcs[0, :], direcs[1, :], direcs[2, :], s=30, c=d[:])
ax.plot([0, u[0]], [0, u[1]], [0, u[2]], c='blue')
ax.set_xlim(-1, 1)
ax.set_ylim(-1, 1)
ax.set_zlim(0, 1)
plt.draw()
count += 1
# Get the istft of the processed data
if PLAY_AUDIO:
new_data = stft.performIStft()
new_data = out_buf.reduce_channels(new_data, NUM_CHANNELS_IN, NUM_CHANNELS_OUT)
# Write out the new, altered data
if out_buf.get_available_write() >= WINDOW_LENGTH:
out_buf.write_samples(new_data)
#time.sleep(.05)
except KeyboardInterrupt:
print "Program interrupted"
done = True
print "Cleaning up"
in_stream.stop_stream()
in_stream.close()
out_stream.stop_stream()
out_stream.close()
pa.terminate()
print "Done"
def localize():
global switch_beamforming
global DO_BEAMFORM
global done
# Setup search space
source_plane = OrientedSourcePlane(SOURCE_PLANE_NORMAL,
SOURCE_PLANE_UP,
SOURCE_PLANE_OFFSET)
space = SearchSpace(MIC_LOC, CAMERA_LOC, [source_plane], MIC_FORWARD, MIC_ABOVE)
# Setup camera
camera = SonyCamera(URL, CAM_FORWARD, CAM_ABOVE)
prev_direc = np.array([1., 0., 0.])
if DO_TRACK:
camera.face_direction(prev_direc) # Will force login
# Setup pyaudio instances
pa = pyaudio.PyAudio()
helper = AudioHelper(pa)
listener = CommandListener()
plot_manager = PlotManager('3d_vm_srp_')
localizer = VonMisesTrackingLocalizer(mic_positions=mic_layout,
search_space=space,
n_particles=N_PARTICLES,
state_kappa=STATE_KAPPA,
#observation_kappa=OBS_KAPPA,
observation_kappa=5,
outlier_prob=.5,
dft_len=FFT_LENGTH,
sample_rate=SAMPLE_RATE,
n_theta=N_THETA,
n_phi=N_PHI)
localizer2 = VonMisesTrackingLocalizer(mic_positions=mic_layout,
search_space=space,
n_particles=N_PARTICLES,
state_kappa=STATE_KAPPA,
#observation_kappa=OBS_KAPPA,
observation_kappa=25,
outlier_prob=0,
dft_len=FFT_LENGTH,
sample_rate=SAMPLE_RATE,
n_theta=N_THETA,
n_phi=N_PHI)
localizer3 = VonMisesTrackingLocalizer(mic_positions=mic_layout,
search_space=space,
n_particles=N_PARTICLES,
state_kappa=STATE_KAPPA,
observation_kappa=OBS_KAPPA,
outlier_prob=.6,
dft_len=FFT_LENGTH,
sample_rate=SAMPLE_RATE,
n_theta=N_THETA,
n_phi=N_PHI)
beamformer = BeamFormer(mic_layout, SAMPLE_RATE)
# Setup STFT object
stft = StftManager(dft_length=FFT_LENGTH,
window_length=WINDOW_LENGTH,
hop_length=HOP_LENGTH,
use_window_fcn=True,
n_channels=NUM_CHANNELS_IN,
dtype=DATA_TYPE)
# Setup devices
in_device = helper.get_input_device_from_user()
if PLAY_AUDIO:
out_device = helper.get_output_device_from_user()
else:
out_device = helper.get_default_output_device_info()
# Setup streams
in_stream = pa.open(rate=SAMPLE_RATE,
channels=NUM_CHANNELS_IN,
format=SAMPLE_TYPE,
frames_per_buffer=FRAMES_PER_BUF,
input=True,
input_device_index=int(in_device['index']),
stream_callback=read_in_data)
out_stream = pa.open(rate=SAMPLE_RATE,
channels=NUM_CHANNELS_OUT,
format=SAMPLE_TYPE,
output=True,
frames_per_buffer=FRAMES_PER_BUF,
output_device_index=int(out_device['index']),
stream_callback=write_out_data)
# Start recording/playing back
in_stream.start_stream()
out_stream.start_stream()
# Start thread to check for user quit
listener.start_polling()
# Setup directions and alignment matrices
direcs = localizer.get_directions()
align_mats = localizer.get_pos_align_mat()
# Plotting setup
if PLOT_PARTICLES:
ml_color = 'r'
color = 'b';
particle_plot = ParticleHemispherePlot(
N_PARTICLES, color, n_estimates=2, n_past_estimates=100,
plot_lines=[False, True], elev=60, azim=45, estim_colors=[ml_color, color])
#color = 'b'
#particle_plot2 = ParticleHemispherePlot(
# N_PARTICLES, color, n_estimates=2, n_past_estimates=100,
# plot_lines=[False, True], elev=60, azim=45, estim_colors=[ml_color, color])
#color = 'r'
#particle_plot3 = ParticleHemispherePlot(
# N_PARTICLES, color, n_estimates=2, n_past_estimates=100,
# plot_lines=[False, True], elev=60, azim=45, estim_colors=[ml_color, color])
if PLOT_POLAR:
fig = plt.figure()
ax = fig.add_subplot(111, projection='polar')
ax.set_rlim(0, 1)
plt.show(block=False)
# Setup space for plotting in new coordinates
spher_coords = localizer.get_spher_directions()
theta = spher_coords[1, :]
pol_plot, = plt.plot(theta, np.ones(theta.shape))
post_plot, = plt. plot(theta, np.ones(theta.shape), 'green')
ax.set_ylim(0, 1)
if DO_BEAMFORM:
pol_beam_plot, = plt.plot(theta, np.ones(theta.shape), 'red')
if PLOT_CARTES:
fig = plt.figure()
ax = fig.add_subplot(111, projection='3d')
plt.show(block=False)
x = localizer.to_spher_grid(direcs[0, :])
y = localizer.to_spher_grid(direcs[1, :])
z = localizer.to_spher_grid(direcs[2, :])
#scat = ax.scatter(x, y, z, s=100)
if EXTERNAL_PLOT:
fig = plt.figure()
ax = fig.add_subplot(111)
plt.show(block=False)
count = 0
estimate = np.array([1., 0., 0.])
estimate2 = np.array([1., 0., 0.])
try:
while in_stream.is_active() or out_stream.is_active():
done = listener.quit()
data_available = in_buf.wait_for_read(WINDOW_LENGTH, TIMEOUT)
if data_available:
if switch_beamforming:
DO_BEAMFORM = not DO_BEAMFORM
switch_beamforming = False
# Get data from the circular buffer
data = in_buf.read_samples(WINDOW_LENGTH)
# Perform an stft
stft.performStft(data)
# Process dfts from windowed segments of input
dfts = stft.getDFTs()
rffts = mat.to_all_real_matlab_format(dfts)
d, energy = localizer.get_distribution_real(rffts[:, :, 0], 'gcc') # Use first hop
# Find ml_est
ml_est = direcs[:, np.argmax(d)]
#print energy
#if energy < 1500:
# continue
post = localizer.get_distribution(rffts[:, :, 0]) # PyBayes EmpPdf
post2 = localizer2.get_distribution(rffts[:, :, 0])
post3 = localizer3.get_distribution(rffts[:, :, 0])
# Get estimate from particles
w = np.asarray(post.weights)
ps = np.asarray(post.particles)
w2 = np.asarray(post2.weights)
ps2 = np.asarray(post2.particles)
w3 = np.asarray(post3.weights)
ps3 = np.asarray(post3.particles)
#estimate2 = w2.dot(ps2)
if DO_TRACK and count % TRACKING_FREQ == 0:
#v = np.array([1, 0, 1])
v = estimate
direc = space.get_camera_dir(v)
if direc is None or not direc.any():
direc = prev_direc
else:
direc[2] = -.5
prev_direc = direc
# Send camera new direction
camera.face_direction(direc)
# Do beam forming
if DO_BEAMFORM:
align_mat = align_mats[:, :, ind]
filtered = beamformer.filter_real(rffts, align_mat)
mat.set_dfts_real(dfts, filtered, n_channels=2)
# Take care of plotting
if count % 1 == 0:
if PLOT_PARTICLES:
estimate = w.dot(ps)
estimate /= (mat.norm2(estimate) + consts.EPS)
particle_plot.update(ps, w, [ml_est, estimate])
#estimate2 = w2.dot(ps2)
#estimate2 /= (mat.norm2(estimate2) + consts.EPS)
#particle_plot2.update(ps2, w2, [ml_est, estimate2])
#estimate3 = w3.dot(ps3)
#estimate3 /= (mat.norm2(estimate3) + consts.EPS)
#particle_plot3.update(ps3, w3, [ml_est, estimate3])
if listener.savefig():
plot_manager.savefig(particle_plot.get_figure())
#plot_manager.savefig(particle_plot2.get_figure())
#plot_manager.savefig(particle_plot3.get_figure())
if PLOT_CARTES:
ax.cla()
ax.grid(False)
#d = localizer.to_spher_grid(post / (np.max(post) + consts.EPS))
#d = localizer.to_spher_grid(d / (np.max(d) + consts.EPS))
ax.scatter(x, y, z, c=d, s=40)
#ax.plot_surface(x, y, z, rstride=1, cstride=1, facecolor=plt.cm.gist_heat(d))
u = estimate
ax.plot([0, u[0]], [0, u[1]], [0, u[2]], c='black', linewidth=3)
if DO_BEAMFORM:
if np.max(np.abs(response)) > 1:
response /= np.max(np.abs(response))
X = response * x
Y = response * y
Z = response * z
ax.plot_surface(X, Y, Z, rstride=1, cstride=1, color='white')
ax.set_xlim(-1, 1)
ax.set_ylim(-1, 1)
ax.set_zlim(0, 1)
#ax.view_init(90, -90)
fig.canvas.draw()
if PLOT_2D:
# Get unconditional distribution
dist = localizer.to_spher_grid(d)
dist -= np.min(dist)
dist /= (np.sum(dist) + consts.EPS)
sample_mat[:, :-1] = sample_mat[:, 1:]
sample_mat[:, -1] = dist
# Get kalman estimate
maxind = np.argmax(post)
estimate_mat[:-1] = estimate_mat[1:]
estimate_mat[-1] = maxind
plot_2d.set_array(sample_mat)
state_est_plot.set_ydata(estimate_mat)
plt.draw()
count += 1
# Get the istft of the processed data
if PLAY_AUDIO or RECORD_AUDIO:
new_data = stft.performIStft()
new_data = out_buf.reduce_channels(new_data, NUM_CHANNELS_IN, NUM_CHANNELS_OUT)
# Write out the new, altered data
if PLAY_AUDIO:
if out_buf.get_available_write() >= WINDOW_LENGTH:
out_buf.write_samples(new_data)
if RECORD_AUDIO:
if record_buf.get_available_write() >= WINDOW_LENGTH:
record_buf.write_samples(new_data)
except KeyboardInterrupt:
print "Program interrupted"
listener.set_quit(True)
print "Cleaning up"
in_stream.stop_stream()
in_stream.close()
out_stream.stop_stream()
out_stream.close()
pa.terminate()
# Take care of output file
if RECORD_AUDIO:
print "Writing output file"
make_wav()
print "Done"