def testTwoChannels(self):
"Test simple stream with two channels"
adapter = FixedSizeInputAdapter(4, 2)
self.assertEquals(len(self.data), adapter.nframes(len(self.data)))
self.assertIOEquals(adapter, self.data[0:1], False, [])
self.assertIOEquals(adapter, self.data[1:5], False, [self.data[0:4]], False)
self.assertIOEquals(adapter, self.data[5:12], False, [self.data[4:8], self.data[8:12]], False)
self.assertIOEquals(adapter, self.data[12:13], False, [])
self.assertIOEquals(adapter, self.data[13:14], False, [])
self.assertIOEquals(adapter, self.data[14:18], False, [self.data[12:16]], False)
self.assertIOEquals(adapter, self.data[18:20], False, [self.data[16:20]], False)
self.assertIOEquals(adapter, self.data[20:21], False, [])
self.assertIOEquals(adapter, self.data[21:22], True, [self.data[20:22]], True)
def testPadding(self):
"Test automatic padding support"
adapter = FixedSizeInputAdapter(4, 2, pad=True)
self.assertEquals(len(self.data) + 2, adapter.nframes(len(self.data)))
self.assertIOEquals(adapter, self.data[0:21], False,
[self.data[0:4], self.data[4:8], self.data[8:12], self.data[12:16], self.data[16:20]],
False)
self.assertIOEquals(adapter, self.data[21:22], True, [[
[20, 42],
[21, 43],
[0, 0],
[0, 0]
]], True)
class SpectrogramImage(object):
""" Builds a PIL image representing a spectrogram of the audio stream (level vs. frequency vs. time).
Adds pixels iteratively thanks to the adapter providing fixed size frame buffers."""
def __init__(self, image_width, image_height, nframes, samplerate, fft_size, bg_color=None, color_scheme='default'):
self.image_width = image_width
self.image_height = image_height
self.nframes = nframes
self.samplerate = samplerate
self.fft_size = fft_size
self.color_scheme = color_scheme
if isinstance(color_scheme, dict):
colors = color_scheme['spectrogram']
else:
colors = default_color_schemes[color_scheme]['spectrogram']
self.image = Image.new("P", (self.image_height, self.image_width))
self.image.putpalette(interpolate_colors(colors, True))
self.samples_per_pixel = self.nframes / float(self.image_width)
self.buffer_size = int(round(self.samples_per_pixel, 0))
self.pixels_adapter = FixedSizeInputAdapter(self.buffer_size, 1, pad=False)
self.pixels_adapter_nframes = self.pixels_adapter.nframes(self.nframes)
self.lower = 100
self.higher = 22050
self.spectrum = Spectrum(self.fft_size, self.nframes, self.samplerate, self.lower, self.higher, numpy.hanning)
# generate the lookup which translates y-coordinate to fft-bin
self.y_to_bin = []
f_min = float(self.lower)
f_max = float(self.higher)
y_min = math.log10(f_min)
y_max = math.log10(f_max)
for y in range(self.image_height):
freq = math.pow(10.0, y_min + y / (image_height - 1.0) *(y_max - y_min))
bin = freq / 22050.0 * (self.fft_size/2 + 1)
if bin < self.fft_size/2:
alpha = bin - int(bin)
self.y_to_bin.append((int(bin), alpha * 255))
# this is a bit strange, but using image.load()[x,y] = ... is
# a lot slower than using image.putadata and then rotating the image
# so we store all the pixels in an array and then create the image when saving
self.pixels = []
self.pixel_cursor = 0
def draw_spectrum(self, x, spectrum):
for (index, alpha) in self.y_to_bin:
self.pixels.append( int( ((255.0-alpha) * spectrum[index] + alpha * spectrum[index + 1] )) )
for y in range(len(self.y_to_bin), self.image_height):
self.pixels.append(0)
def process(self, frames, eod):
if len(frames) != 1:
buffer = frames[:,0].copy()
buffer.shape = (len(buffer),1)
# FIXME : breaks spectrum linearity
for samples, end in self.pixels_adapter.process(buffer, eod):
if self.pixel_cursor < self.image_width:
(spectral_centroid, db_spectrum) = self.spectrum.process(samples, True)
self.draw_spectrum(self.pixel_cursor, db_spectrum)
self.pixel_cursor += 1
def watermark(self, text, color=None, opacity=.6, margin=(10,10)):
#self.image = im_watermark(self.image, text, color=color, opacity=opacity, margin=margin)
pass
def save(self, filename):
""" Apply last 2D transforms and write all pixels to the file. """
self.image.putdata(self.pixels)
self.image.transpose(Image.ROTATE_90).save(filename)
def release(self):
pass
class WaveformImageSimple(object):
""" Builds a PIL image representing a waveform of the audio stream.
Adds pixels iteratively thanks to the adapter providing fixed size frame buffers.
"""
def __init__(self, image_width, image_height, nframes, samplerate, fft_size, bg_color, color_scheme):
self.image_width = image_width
self.image_height = image_height
self.nframes = nframes
self.samplerate = samplerate
self.fft_size = fft_size
self.bg_color = bg_color
self.color_scheme = color_scheme
if isinstance(color_scheme, dict):
colors = color_scheme['waveform']
else:
colors = default_color_schemes[color_scheme]['waveform']
self.line_color = colors[0]
self.samples_per_pixel = self.nframes / float(self.image_width)
self.buffer_size = int(round(self.samples_per_pixel, 0))
self.pixels_adapter = FixedSizeInputAdapter(self.buffer_size, 1, pad=False)
self.pixels_adapter_nframes = self.pixels_adapter.nframes(self.nframes)
self.image = Image.new("RGBA", (self.image_width, self.image_height))
self.pixel = self.image.load()
self.draw = ImageDraw.Draw(self.image)
self.previous_x, self.previous_y = None, None
self.frame_cursor = 0
self.pixel_cursor = 0
def normalize(self, contour):
contour = contour-min(contour)
return contour/max(contour)
def peaks(self, samples):
""" Find the minimum and maximum peak of the samples.
Returns that pair in the order they were found.
So if min was found first, it returns (min, max) else the other way around. """
max_index = numpy.argmax(samples)
max_value = samples[max_index]
min_index = numpy.argmin(samples)
min_value = samples[min_index]
if min_index < max_index:
return (min_value, max_value)
else:
return (max_value, min_value)
def draw_peaks(self, x, peaks):
""" draw 2 peaks at x using the spectral_centroid for color """
y1 = self.image_height * 0.5 - peaks[0] * (self.image_height - 4) * 0.5
y2 = self.image_height * 0.5 - peaks[1] * (self.image_height - 4) * 0.5
if self.previous_y and x < self.image_width-1:
if y1 < y2:
self.draw.line((x, 0, x, y1), self.line_color)
self.draw.line((x, self.image_height , x, y2), self.line_color)
else:
self.draw.line((x, 0, x, y2), self.line_color)
self.draw.line((x, self.image_height , x, y1), self.line_color)
else:
self.draw.line((x, 0, x, self.image_height), self.line_color)
self.previous_x, self.previous_y = x, y1
def process(self, frames, eod):
if len(frames) != 1:
buffer = frames[:,0]
buffer.shape = (len(buffer),1)
for samples, end in self.pixels_adapter.process(buffer, eod):
if self.pixel_cursor < self.image_width-1:
self.draw_peaks(self.pixel_cursor, self.peaks(samples))
self.pixel_cursor += 1
if end:
samples = 0
buffer = 0
break
if self.pixel_cursor == self.image_width-1:
self.draw_peaks(self.pixel_cursor, (0, 0))
self.pixel_cursor += 1
def watermark(self, text, color=None, opacity=.6, margin=(10,10)):
self.image = im_watermark(self.image, text, color=color, opacity=opacity, margin=margin)
def save(self, filename):
""" Apply last 2D transforms and write all pixels to the file. """
# middle line (0 for none)
a = 1
for x in range(self.image_width):
self.pixel[x, self.image_height/2] = tuple(map(lambda p: p+a, self.pixel[x, self.image_height/2]))
self.image.save(filename)
def release(self):
pass
class WaveformImage(object):
""" Builds a PIL image representing a waveform of the audio stream.
Adds pixels iteratively thanks to the adapter providing fixed size frame buffers.
Peaks are colored relative to the spectral centroids of each frame packet. """
def __init__(self, image_width, image_height, nframes, samplerate, fft_size, bg_color, color_scheme):
self.image_width = image_width
self.image_height = image_height
self.nframes = nframes
self.samplerate = samplerate
self.fft_size = fft_size
self.bg_color = bg_color
self.color_scheme = color_scheme
if isinstance(color_scheme, dict):
colors = color_scheme['waveform']
else:
colors = default_color_schemes[color_scheme]['waveform']
self.color_lookup = interpolate_colors(colors)
self.samples_per_pixel = self.nframes / float(self.image_width)
self.buffer_size = int(round(self.samples_per_pixel, 0))
self.pixels_adapter = FixedSizeInputAdapter(self.buffer_size, 1, pad=False)
self.pixels_adapter_nframes = self.pixels_adapter.nframes(self.nframes)
self.lower = 800
self.higher = 12000
self.spectrum = Spectrum(self.fft_size, self.nframes, self.samplerate, self.lower, self.higher, numpy.hanning)
self.image = Image.new("RGBA", (self.image_width, self.image_height), self.bg_color)
self.pixel = self.image.load()
self.draw = ImageDraw.Draw(self.image)
self.previous_x, self.previous_y = None, None
self.frame_cursor = 0
self.pixel_cursor = 0
def peaks(self, samples):
""" Find the minimum and maximum peak of the samples.
Returns that pair in the order they were found.
So if min was found first, it returns (min, max) else the other way around. """
max_index = numpy.argmax(samples)
max_value = samples[max_index]
min_index = numpy.argmin(samples)
min_value = samples[min_index]
if min_index < max_index:
return (min_value, max_value)
else:
return (max_value, min_value)
def color_from_value(self, value):
""" given a value between 0 and 1, return an (r,g,b) tuple """
return ImageColor.getrgb("hsl(%d,%d%%,%d%%)" % (int( (1.0 - value) * 360 ), 80, 50))
def draw_peaks(self, x, peaks, spectral_centroid):
""" draw 2 peaks at x using the spectral_centroid for color """
y1 = self.image_height * 0.5 - peaks[0] * (self.image_height - 4) * 0.5
y2 = self.image_height * 0.5 - peaks[1] * (self.image_height - 4) * 0.5
line_color = self.color_lookup[int(spectral_centroid*255.0)]
if self.previous_y:
self.draw.line([self.previous_x, self.previous_y, x, y1, x, y2], line_color)
else:
self.draw.line([x, y1, x, y2], line_color)
self.previous_x, self.previous_y = x, y2
self.draw_anti_aliased_pixels(x, y1, y2, line_color)
def draw_anti_aliased_pixels(self, x, y1, y2, color):
""" vertical anti-aliasing at y1 and y2 """
y_max = max(y1, y2)
y_max_int = int(y_max)
alpha = y_max - y_max_int
if alpha > 0.0 and alpha < 1.0 and y_max_int + 1 < self.image_height:
current_pix = self.pixel[int(x), y_max_int + 1]
r = int((1-alpha)*current_pix[0] + alpha*color[0])
g = int((1-alpha)*current_pix[1] + alpha*color[1])
b = int((1-alpha)*current_pix[2] + alpha*color[2])
self.pixel[x, y_max_int + 1] = (r,g,b)
y_min = min(y1, y2)
y_min_int = int(y_min)
alpha = 1.0 - (y_min - y_min_int)
if alpha > 0.0 and alpha < 1.0 and y_min_int - 1 >= 0:
current_pix = self.pixel[x, y_min_int - 1]
r = int((1-alpha)*current_pix[0] + alpha*color[0])
g = int((1-alpha)*current_pix[1] + alpha*color[1])
b = int((1-alpha)*current_pix[2] + alpha*color[2])
self.pixel[x, y_min_int - 1] = (r,g,b)
def process(self, frames, eod):
if len(frames) != 1:
buffer = frames[:,0].copy()
buffer.shape = (len(buffer),1)
(spectral_centroid, db_spectrum) = self.spectrum.process(buffer, True)
for samples, end in self.pixels_adapter.process(buffer, eod):
if self.pixel_cursor < self.image_width:
peaks = self.peaks(samples)
self.draw_peaks(self.pixel_cursor, peaks, spectral_centroid)
self.pixel_cursor += 1
def watermark(self, text, color=None, opacity=.6, margin=(10,10)):
self.image = im_watermark(self.image, text, color=color, opacity=opacity, margin=margin)
def save(self, filename):
""" Apply last 2D transforms and write all pixels to the file. """
# middle line (0 for none)
a = 1
for x in range(self.image_width):
self.pixel[x, self.image_height/2] = tuple(map(lambda p: p+a, self.pixel[x, self.image_height/2]))
self.image.save(filename)
def release(self):
pass