def signal_test (small_exponent, large_exponent, signal, filename):
fft_size = 1 << ((small_exponent + large_exponent) / 2)
length_windows = fft_size / 2
length_samples = fft_size * length_windows
print "creating signal"
input = K.Complexes(length_samples)
for i in range(length_samples):
t = (0.5 + i) / length_samples
input[i] = signal(t)
print "transforming signal"
s = K.Multigram(small_exponent, large_exponent)
length_frames = length_samples / s.size_in
input = input.reshape((length_frames, s.size_in))
output = K.Reals(length_frames, s.size_out)
for i in range(length_frames):
s.transform_fwd(input[i,:], output[i,:])
print "writing multigram to %s" % filename
output = util.energy_to_loudness(output)
formats.write_image(output, filename)
def show_signal (size_exponent, factor_exponent, signal, filename):
size = 1 << size_exponent
factor = 1 << factor_exponent
length_windows = size / 2
length_samples = size * length_windows
print "creating signal"
input = K.transforms.Complexes(length_samples)
for i in range(length_samples):
t = (0.5 + i) / length_samples
input[i] = signal(t)
#print signal(t), #DEBUG
print "transforming signal"
s = K.Supergram(size_exponent, factor_exponent)
length_frames = length_samples / s.small_size
input = input.reshape((length_frames, s.small_size))
output = K.transforms.Reals(length_frames, s.super_size)
for i in range(length_frames):
s.transform_fwd(input[i,:], output[i,:])
print "writing supergram to %s" % filename
#XXX this fails due to NaNs and Infs in output XXX
#output[~isfinite(output)] = 0 #DEBUG
output = K.util.energy_to_loudness(output, 0.1)
formats.write_image(output, filename).show()
def show_signal(bank_size, block_size, signal):
size_in = block_size
size_out = bank_size
num_blocks = bank_size * 16 / 9 # arbitrary
duration = num_blocks * block_size
min_freq = 0.25 / block_size
max_freq = 0.25
pitchgram = K.Pitchgram(size_in, size_out, min_freq, max_freq)
print "creating signal"
t = array(range(duration)) / block_size
sound = K.transforms.Complexes(duration)
sound[:] = signal(t)
sound = sound.reshape((num_blocks, block_size))
print "transforming signal"
image = K.transforms.Reals(num_blocks, size_out)
for i in range(num_blocks):
pitchgram.transform(sound[i, :], image[i, :])
print "writing pitchgram to pitchgram.png"
# XXX this fails due to NaNs and Infs in image XXX
# image[~isfinite(image)] = 0 #DEBUG
image = K.util.energy_to_loudness(image, 0.1)
formats.write_image(image, "pitchgram.png").show()
def file_test (small_exponent = 9, large_exponent = 12):
"multigram"
s = K.Multigram(small_exponent, large_exponent)
size_in = s.size_in
large_size = s.size_out
length = s.size_out / 2#arbitrary
print "reading sound file"
sound = formats.read_wav('test.wav', size_in*length, size_in)
image = K.Reals(length, large_size)
print "transforming data"
for i in range(length):
s.transform_fwd(sound[i,:], image[i,:])
del sound
print "saving image"
image = util.energy_to_loudness(image + 1e-5)
formats.write_image(image, 'test.png')
def show_file(bank_size=512, block_size=K.DEFAULT_FRAMES_PER_BUFFER):
"Shows pitchgram of an audio file"
size_in = block_size
size_out = bank_size
length = size_out * 16 / 9 # arbitrary
pitchgram = K.Pitchgram(size_in, size_out)
print "reading sound file"
sound = formats.read_wav("test.wav", size_in * length, size_in)
image = K.transforms.Reals(length, size_out)
print "transforming data"
for i in range(length):
pitchgram.transform(sound[i, :], image[i, :])
# del sound
print "saving image"
image = K.util.energy_to_loudness(image + 1e-5)
formats.write_image(image, "test.png").show()
def show_file (size_exponent = 10, time_exponent = 2, freq_exponent = 3):
'Shows supergram of an audio file'
s = K.Supergram(size_exponent, time_exponent, freq_exponent)
size = s.size
small_size = s.small_size
large_size = s.super_size
length = s.super_size #arbitrary
print "reading sound file"
sound = formats.read_wav('test.wav', small_size*length, small_size)
image = K.transforms.Reals(length, large_size)
print "transforming data"
for i in range(length):
s.transform_fwd(sound[i,:], image[i,:])
#del sound
print "saving image"
image = K.util.energy_to_loudness(image + 1e-5)
formats.write_image(image, 'test.png').show()
def signal_test (exponent, signal, filename):
width = 1 << exponent
size = width * width
print "creating signal"
input = K.Complexes(size)
for i in range(size):
t = (i + 0.5) / size
input[i] = signal(t)
input = input.reshape((width,width))
print "transforming signal"
output = K.Reals(width,width/2)
s = K.Spectrogram(exponent)
for i in range(width):
s.transform_fwd(input[i,:], output[i,:])
print "writing spactrogram to %s" % filename
output = util.energy_to_loudness(output)
formats.write_image(output, filename)
def read_spectrogram (exponent = 10, wavname = 'test.wav'):
"""
read a sound file, say a wav file;
compute a spectrogram or reassigned spectrogram;
write image to png file
"""
width = 1 << exponent
height = width
size = height * width
print "reading sound file"
sound = formats.read_wav(wavname, size, width)
image = K.Reals(height, width/2)
print "transforming data"
s = K.Spectrogram(exponent)
for i in range(height):
s.transform_fwd(sound[i,:],image[i,:])
print "saving image"
image = formats.energy_to_loudness(image)
formats.write_image(image, 'test.png')
def record_spectrogram(exponent=10):
width = 1 << exponent
height = width
size = height * width
sound = K.Complexes(width)
image = K.Reals(height, width / 2)
s = K.Spectrogram(exponent)
a = K.Audio(width)
time = size * a.rate / 60.0
print "recording & transforming sound for %g seconds..." % time
a.start()
for i in range(height):
a.read(sound)
s.transform_fwd(sound, image[i, :])
a.write(sound) # HACK
a.stop()
print "saving image"
image = formats.energy_to_loudness(image)
formats.write_image(image, "test.png")