def localize():
# Global variables that may be set in this function
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 pyaudio instances
pa = pyaudio.PyAudio()
helper = AudioHelper(pa)
listener = CommandListener()
plot_manager = PlotManager()
#localizer = GridTrackingLocalizer(mic_positions=mic_layout,
# search_space=space,
# source_cov=SOURCE_LOCATION_COV,
# dft_len=FFT_LENGTH,
# sample_rate=SAMPLE_RATE,
# n_theta=N_THETA,
# n_phi=N_PHI)
localizer = KalmanTrackingLocalizer(mic_positions=mic_layout,
search_space=space,
mic_forward=MIC_FORWARD,
mic_above=MIC_ABOVE,
trans_mat=STATE_TRANSITION_MAT,
state_cov=STATE_TRANSITION_MAT,
emission_mat=EMISSION_MAT,
emission_cov=EMISSION_COV,
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_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.zeros(theta.shape))
post_plot, = plt. plot(theta, np.zeros(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 = plotting.get_halfpage_axis(fig)
#ax = fig.add_subplot(111)
plt.show(block=False)
# Setup space for plotting in new coordinates
spher_coords = localizer.get_spher_directions()
theta = spher_coords[1, :]
theta = np.linspace(0, 1, theta.shape[0])
gcc_plots = []
gcc_shaping_vals = [1, 2, 3, 4, 5]
for i in gcc_shaping_vals:
plot, = plt.plot(theta, np.zeros(theta.shape))
gcc_plots.append(plot)
pol_plot, = plt.plot(theta, np.zeros(theta.shape), 'r--')
post_plot, = plt. plot(theta, np.zeros(theta.shape), 'b')
ax.set_ylim(0, 1.2)
ax.set_xlim(0, 1) # Normalized
#ax.set_xlabel('Angle $\left(\\frac{1}{\pi}\\right)$')
#ax.set_ylabel('Normalized GCC')
if DO_BEAMFORM:
pol_beam_plot, = plt.plot(theta, np.ones(theta.shape), 'red')
if PLOT_2D:
n_past_samples = 200
filter_plot = FilterPlot(N_THETA, n_past_samples, 2)
save_plot = filter_plot # For saving figures
if VIDEO_OVERLAY:
vc = cv2.VideoCapture(0)
video_handle, video_plot = setup_video_handle(720, 1280)
plt.show(block=False)
if EXTERNAL_PLOT:
fig = plt.figure()
ax = fig.add_subplot(111)
plt.show(block=False)
count = 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 listener.switch_beamforming():
DO_BEAMFORM = not DO_BEAMFORM
if listener.savefig():
plot_manager.savefig(save_plot.get_figure())
# Get data from 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)
gccs = []
#for k in gcc_shaping_vals:
# d, energy = localizer.get_distribution_real(
# rffts[:, :, 0], 'mcc', k) # Use first hop
# gccs.append(d)
d, energy = localizer.get_distribution_real(rffts[:, :, 0], 'beam') # Use first hop
def w(cpmat):
cpmat /= (np.abs(cpmat + consts.EPS))
return cpmat
post = localizer.get_distribution(rffts[:, :, 0], 'beam')
#post, bla = localizer.get_distribution_real(rffts[:, :, 0], 'mcc')
#post = localizer.get_distribution(rffts[:, :, 0])
ind = np.argmax(d)
u = 1.5 * direcs[:, ind] # Direction of arrival
#if energy < 500:
#continue
# 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_POLAR or PLOT_CARTES:
dist = d
#dist -= np.min(dist)
dist = localizer.to_spher_grid(dist)
print post.shape
post = localizer.to_spher_grid(post) * 50
dist /= np.max(dist)
if np.max(dist) > 1:
dist /= np.max(dist)
if np.max(post) > 1:
post /= np.max(post)
pol_plot.set_ydata(dist[0, :])
post_plot.set_ydata(post[0, :])
#for i, plot in enumerate(gcc_plots):
# gcc = gccs[i]
# gcc /= (np.max(gcc) + consts.EPS)
# plot.set_ydata(gccs[i])
if DO_BEAMFORM:
# Get beam plot
freq = 2500. # Hz
response = beamformer.get_beam(
align_mat, align_mats, rffts, freq
)
response = localizer.to_spher_grid(response)
if np.max(response) > 1:
response /= np.max(response)
pol_beam_plot.set_ydata(response[-1, :])
plt.draw()
if PLOT_2D:
dist = localizer.to_spher_grid(d)
p = localizer.to_spher_grid(post)
est1 = THETA_SPACE[np.argmax(p)]
est2 = THETA_SPACE[np.argmax(dist)]
filter_plot.update(dist, [est1, est2])
if VIDEO_OVERLAY:
post /= np.max(post + consts.EPS)
dist = d - np.min(d)
dist = dist / np.max(dist + consts.EPS)
_, cvimage = vc.read()
overlay_distribution(video_handle, video_plot, cvimage, dist[::-1])
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"
def localize():
global switch_beamforming
global DO_BEAMFORM
# Setup search space
source_plane = OrientedSourcePlane(SOURCE_PLANE_NORMAL,
SOURCE_PLANE_UP,
SOURCE_PLANE_OFFSET)
space = SearchSpace(MIC_LOC, CAMERA_LOC, [source_plane])
# Setup pyaudio instances
pa = pyaudio.PyAudio()
helper = AudioHelper(pa)
localizer = KalmanTrackingLocalizer(mic_positions=mic_layout,
search_space=space,
mic_forward=MIC_FORWARD,
mic_above=MIC_ABOVE,
trans_mat=STATE_TRANSITION_MAT,
state_cov=STATE_TRANSITION_MAT,
emission_mat=EMISSION_MAT,
emission_cov=EMISSION_COV,
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
quit_thread = threading.Thread(target=check_for_quit)
quit_thread.start()
# Setup directions and alignment matrices
direcs = localizer.get_directions()
align_mats = localizer.get_pos_align_mat()
# Plotting setup
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
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:
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
post = localizer.get_distribution(rffts[:, :, 0])
ind = np.argmax(post)
u = 1.5 * direcs[:, ind] # Direction of arrival
#if energy < 500:
# continue
# 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_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))
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"
done = 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"
def localize():
# Global variables that may be set in this function
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 pyaudio instances
pa = pyaudio.PyAudio()
helper = AudioHelper(pa)
listener = CommandListener()
plot_manager = PlotManager()
#localizer = GridTrackingLocalizer(mic_positions=mic_layout,
# search_space=space,
# source_cov=SOURCE_LOCATION_COV,
# dft_len=FFT_LENGTH,
# sample_rate=SAMPLE_RATE,
# n_theta=N_THETA,
# n_phi=N_PHI)
localizer = KalmanTrackingLocalizer(mic_positions=mic_layout,
search_space=space,
mic_forward=MIC_FORWARD,
mic_above=MIC_ABOVE,
trans_mat=STATE_TRANSITION_MAT,
state_cov=STATE_TRANSITION_MAT,
emission_mat=EMISSION_MAT,
emission_cov=EMISSION_COV,
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_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.zeros(theta.shape))
post_plot, = plt.plot(theta, np.zeros(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 = plotting.get_halfpage_axis(fig)
#ax = fig.add_subplot(111)
plt.show(block=False)
# Setup space for plotting in new coordinates
spher_coords = localizer.get_spher_directions()
theta = spher_coords[1, :]
theta = np.linspace(0, 1, theta.shape[0])
gcc_plots = []
gcc_shaping_vals = [1, 2, 3, 4, 5]
for i in gcc_shaping_vals:
plot, = plt.plot(theta, np.zeros(theta.shape))
gcc_plots.append(plot)
pol_plot, = plt.plot(theta, np.zeros(theta.shape), 'r--')
post_plot, = plt.plot(theta, np.zeros(theta.shape), 'b')
ax.set_ylim(0, 1.2)
ax.set_xlim(0, 1) # Normalized
#ax.set_xlabel('Angle $\left(\\frac{1}{\pi}\\right)$')
#ax.set_ylabel('Normalized GCC')
if DO_BEAMFORM:
pol_beam_plot, = plt.plot(theta, np.ones(theta.shape), 'red')
if PLOT_2D:
n_past_samples = 200
filter_plot = FilterPlot(N_THETA, n_past_samples, 2)
save_plot = filter_plot # For saving figures
if VIDEO_OVERLAY:
vc = cv2.VideoCapture(0)
video_handle, video_plot = setup_video_handle(720, 1280)
plt.show(block=False)
if EXTERNAL_PLOT:
fig = plt.figure()
ax = fig.add_subplot(111)
plt.show(block=False)
count = 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 listener.switch_beamforming():
DO_BEAMFORM = not DO_BEAMFORM
if listener.savefig():
plot_manager.savefig(save_plot.get_figure())
# Get data from 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)
gccs = []
#for k in gcc_shaping_vals:
# d, energy = localizer.get_distribution_real(
# rffts[:, :, 0], 'mcc', k) # Use first hop
# gccs.append(d)
d, energy = localizer.get_distribution_real(
rffts[:, :, 0], 'beam') # Use first hop
def w(cpmat):
cpmat /= (np.abs(cpmat + consts.EPS))
return cpmat
post = localizer.get_distribution(rffts[:, :, 0], 'beam')
#post, bla = localizer.get_distribution_real(rffts[:, :, 0], 'mcc')
#post = localizer.get_distribution(rffts[:, :, 0])
ind = np.argmax(d)
u = 1.5 * direcs[:, ind] # Direction of arrival
#if energy < 500:
#continue
# 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_POLAR or PLOT_CARTES:
dist = d
#dist -= np.min(dist)
dist = localizer.to_spher_grid(dist)
print post.shape
post = localizer.to_spher_grid(post) * 50
dist /= np.max(dist)
if np.max(dist) > 1:
dist /= np.max(dist)
if np.max(post) > 1:
post /= np.max(post)
pol_plot.set_ydata(dist[0, :])
post_plot.set_ydata(post[0, :])
#for i, plot in enumerate(gcc_plots):
# gcc = gccs[i]
# gcc /= (np.max(gcc) + consts.EPS)
# plot.set_ydata(gccs[i])
if DO_BEAMFORM:
# Get beam plot
freq = 2500. # Hz
response = beamformer.get_beam(
align_mat, align_mats, rffts, freq)
response = localizer.to_spher_grid(response)
if np.max(response) > 1:
response /= np.max(response)
pol_beam_plot.set_ydata(response[-1, :])
plt.draw()
if PLOT_2D:
dist = localizer.to_spher_grid(d)
p = localizer.to_spher_grid(post)
est1 = THETA_SPACE[np.argmax(p)]
est2 = THETA_SPACE[np.argmax(dist)]
filter_plot.update(dist, [est1, est2])
if VIDEO_OVERLAY:
post /= np.max(post + consts.EPS)
dist = d - np.min(d)
dist = dist / np.max(dist + consts.EPS)
_, cvimage = vc.read()
overlay_distribution(video_handle, video_plot, cvimage,
dist[::-1])
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"
def localize():
global switch_beamforming
global DO_BEAMFORM
# Setup search space
source_plane = OrientedSourcePlane(SOURCE_PLANE_NORMAL, SOURCE_PLANE_UP,
SOURCE_PLANE_OFFSET)
space = SearchSpace(MIC_LOC, CAMERA_LOC, [source_plane])
# Setup pyaudio instances
pa = pyaudio.PyAudio()
helper = AudioHelper(pa)
localizer = KalmanTrackingLocalizer(mic_positions=mic_layout,
search_space=space,
mic_forward=MIC_FORWARD,
mic_above=MIC_ABOVE,
trans_mat=STATE_TRANSITION_MAT,
state_cov=STATE_TRANSITION_MAT,
emission_mat=EMISSION_MAT,
emission_cov=EMISSION_COV,
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
quit_thread = threading.Thread(target=check_for_quit)
quit_thread.start()
# Setup directions and alignment matrices
direcs = localizer.get_directions()
align_mats = localizer.get_pos_align_mat()
# Plotting setup
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
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:
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
post = localizer.get_distribution(rffts[:, :, 0])
ind = np.argmax(post)
u = 1.5 * direcs[:, ind] # Direction of arrival
#if energy < 500:
# continue
# 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_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))
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"
done = 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"