class Spectrogram_Widget(QtWidgets.QWidget):
name = "Spectrogram"
def __init__(self, parent, logger=PrintLogger()):
super().__init__(parent)
self.logger = logger
self.setObjectName(self.name)
self.gridLayout = QtWidgets.QGridLayout(self)
self.gridLayout.setObjectName("gridLayout")
self.PlotZoneImage = ImagePlot(self, self.logger)
self.PlotZoneImage.setObjectName("PlotZoneImage")
self.gridLayout.addWidget(self.PlotZoneImage, 0, 1, 1, 1)
# initialize the class instance that will do the fft
self.proc = audioproc(self.logger)
self.maxfreq = DEFAULT_MAXFREQ
self.proc.set_maxfreq(self.maxfreq)
self.minfreq = DEFAULT_MINFREQ
self.fft_size = 2 ** DEFAULT_FFT_SIZE * 32
self.proc.set_fftsize(self.fft_size)
self.spec_min = DEFAULT_SPEC_MIN
self.spec_max = DEFAULT_SPEC_MAX
self.weighting = DEFAULT_WEIGHTING
self.update_weighting()
self.freq = self.proc.get_freq_scale()
self.timerange_s = DEFAULT_TIMERANGE
self.canvas_width = 100.
self.overlap = 3. / 4.
self.overlap_frac = Fraction(3, 4)
self.dT_s = self.fft_size * (1. - self.overlap) / float(SAMPLING_RATE)
self.data_buffer = RingBuffer(1, 4 * self.fft_size)
self.smoothing_buffer = RingBuffer(1, self.fft_size * self.overlap)
self.PlotZoneImage.setlog10freqscale() # DEFAULT_FREQ_SCALE = 1 #log10
self.PlotZoneImage.setfreqrange(self.minfreq, self.maxfreq)
self.PlotZoneImage.setspecrange(self.spec_min, self.spec_max)
self.PlotZoneImage.setweighting(self.weighting)
self.PlotZoneImage.settimerange(self.timerange_s, self.dT_s)
self.update_jitter()
sfft_rate_frac = Fraction(SAMPLING_RATE, self.fft_size) / (Fraction(1) - self.overlap_frac) / 1000
self.PlotZoneImage.set_sfft_rate(sfft_rate_frac)
# initialize the settings dialog
self.settings_dialog = Spectrogram_Settings_Dialog(self, self.logger)
@staticmethod
def log_spectrogram(sp):
# Note: implementing the log10 of the array in Cython did not bring
# any speedup.
# Idea: Instead of computing the log of the data, I could pre-compute
# a list of values associated with the colormap, and then do a search...
epsilon = 1e-30
return 10. * log10(sp + epsilon)
# scale the db spectrum from [- spec_range db ... 0 db] to [0..1] (do not clip, will be down after resampling)
def scale_spectrogram(self, sp):
return (sp - self.spec_min) / (self.spec_max - self.spec_min)
def handle_new_data(self, floatdata):
self.data_buffer.push(floatdata)
new_sample_points = self.data_buffer.num_unread_data_points()
# if we have enough data to add a frequency column in the time-frequency plane, compute it
needed = int(self.fft_size * (1. - self.overlap))
realizable = int(floor(new_sample_points / needed))
if realizable > 0:
spn = zeros((len(self.freq), realizable), dtype=float64)
data = append(self.smoothing_buffer.unwound_data(), self.data_buffer.pop(realizable * needed))
# append flattens 1d arrays, so reshape if necessary
if data.size == data.shape[0]:
data = data.reshape([1, data.size])
for i in range(realizable):
floatdata = data[:, i * needed:i * needed + self.fft_size]
self.smoothing_buffer.push(floatdata[:, (self.fft_size - needed):])
# for now, take the first channel only
floatdata = floatdata[0, :]
# FFT transform
spn[:, i] = self.proc.analyzelive(floatdata, "Power")
w = tile(self.w, (1, realizable))
norm_spectrogram = self.scale_spectrogram(self.log_spectrogram(spn) + w)
self.PlotZoneImage.addData(self.freq, norm_spectrogram)
# thickness of a frequency column depends on FFT size and window overlap
# hamming window with 75% overlap provides good quality (Perfect reconstruction,
# aliasing from side lobes only, 42 dB channel isolation)
# number of frequency columns that we keep depends on the time history that the user has chosen
# actual displayed spectrogram is a scaled version of the time-frequency plane
def canvasUpdate(self):
if not self.isVisible():
return
self.PlotZoneImage.draw()
def update_jitter(self):
audio_jitter = 2 * float(FRAMES_PER_BUFFER) / SAMPLING_RATE
analysis_jitter = self.fft_size * (1. - self.overlap) / SAMPLING_RATE
canvas_jitter = audio_jitter + analysis_jitter
# print audio_jitter, analysis_jitter, canvas_jitter
self.PlotZoneImage.plotImage.set_jitter(canvas_jitter)
def pause(self):
pass
def setminfreq(self, freq):
self.minfreq = freq
self.PlotZoneImage.setfreqrange(self.minfreq, self.maxfreq)
def setmaxfreq(self, freq):
self.maxfreq = freq
self.PlotZoneImage.setfreqrange(self.minfreq, self.maxfreq)
self.proc.set_maxfreq(freq)
self.update_weighting()
self.freq = self.proc.get_freq_scale()
def setfftsize(self, fft_size):
self.fft_size = fft_size
self.proc.set_fftsize(fft_size)
self.update_weighting()
self.freq = self.proc.get_freq_scale()
self.dT_s = self.fft_size * (1. - self.overlap) / float(SAMPLING_RATE)
self.PlotZoneImage.settimerange(self.timerange_s, self.dT_s)
sfft_rate_frac = Fraction(SAMPLING_RATE, self.fft_size) / (Fraction(1) - self.overlap_frac) / 1000
self.PlotZoneImage.set_sfft_rate(sfft_rate_frac)
self.update_jitter()
def set_spect_min(self, value):
self.spec_min = value
self.PlotZoneImage.setspecrange(self.spec_min, self.spec_max)
def set_spect_max(self, value):
self.spec_max = value
self.PlotZoneImage.setspecrange(self.spec_min, self.spec_max)
def setweighting(self, weighting):
self.weighting = weighting
self.PlotZoneImage.setweighting(weighting)
self.update_weighting()
def update_weighting(self):
A, B, C = self.proc.get_freq_weighting()
if self.weighting is 0:
self.w = array([0.])
elif self.weighting is 1:
self.w = A
elif self.weighting is 2:
self.w = B
else:
self.w = C
self.w.shape = (len(self.w), 1)
def settings_called(self, checked):
self.settings_dialog.show()
def saveState(self, settings):
self.settings_dialog.saveState(settings)
def restoreState(self, settings):
self.settings_dialog.restoreState(settings)
# slot
def timerangechanged(self, value):
self.timerange_s = value
self.PlotZoneImage.settimerange(self.timerange_s, self.dT_s)
# slot
def canvasWidthChanged(self, width):
self.canvas_width = width
class Spectrogram_Widget(QtGui.QWidget):
def __init__(self, parent, audiobackend, logger = PrintLogger()):
QtGui.QWidget.__init__(self, parent)
self.logger = logger
self.setObjectName("Spectrogram_Widget")
self.gridLayout = QtGui.QGridLayout(self)
self.gridLayout.setObjectName("gridLayout")
self.PlotZoneImage = ImagePlot(self, self.logger, audiobackend)
#self.PlotZoneImage = GLRollingCanvasWidget(self, self.logger)
self.PlotZoneImage.setObjectName("PlotZoneImage")
self.gridLayout.addWidget(self.PlotZoneImage, 0, 1, 1, 1)
self.audiobuffer = None
self.audiobackend = audiobackend
# initialize the class instance that will do the fft
self.proc = audioproc(self.logger)
self.maxfreq = DEFAULT_MAXFREQ
self.proc.set_maxfreq(self.maxfreq)
self.minfreq = DEFAULT_MINFREQ
self.fft_size = 2**DEFAULT_FFT_SIZE*32
self.proc.set_fftsize(self.fft_size)
self.spec_min = DEFAULT_SPEC_MIN
self.spec_max = DEFAULT_SPEC_MAX
self.weighting = DEFAULT_WEIGHTING
self.update_weighting()
self.freq = self.proc.get_freq_scale()
self.timerange_s = DEFAULT_TIMERANGE
self.canvas_width = 100.
self.old_index = 0
self.overlap = 3./4.
self.overlap_frac = Fraction(3, 4)
self.dT_s = self.fft_size*(1. - self.overlap)/float(SAMPLING_RATE)
self.PlotZoneImage.setlog10freqscale() #DEFAULT_FREQ_SCALE = 1 #log10
self.PlotZoneImage.setfreqrange(self.minfreq, self.maxfreq)
self.PlotZoneImage.setspecrange(self.spec_min, self.spec_max)
self.PlotZoneImage.setweighting(self.weighting)
self.PlotZoneImage.settimerange(self.timerange_s, self.dT_s)
sfft_rate_frac = Fraction(SAMPLING_RATE, self.fft_size)/(Fraction(1) - self.overlap_frac)/1000
self.PlotZoneImage.set_sfft_rate(sfft_rate_frac)
# initialize the settings dialog
self.settings_dialog = Spectrogram_Settings_Dialog(self, self.logger)
# method
def set_buffer(self, buffer):
self.audiobuffer = buffer
self.old_index = self.audiobuffer.ringbuffer.offset
def log_spectrogram(self, sp):
# Note: implementing the log10 of the array in Cython did not bring
# any speedup.
# Idea: Instead of computing the log of the data, I could pre-compute
# a list of values associated with the colormap, and then do a search...
epsilon = 1e-30
return 10.*log10(sp + epsilon)
# scale the db spectrum from [- spec_range db ... 0 db] > [0..1]
def scale_spectrogram(self, sp):
return (sp.clip(min = self.spec_min, max = self.spec_max) - self.spec_min)/(self.spec_max - self.spec_min)
# method
def update(self):
if not self.isVisible():
return
# we need to maintain an index of where we are in the buffer
index = self.audiobuffer.ringbuffer.offset
available = index - self.old_index
if available < 0:
#ringbuffer must have grown or something...
available = 0
self.old_index = index
# if we have enough data to add a frequency column in the time-frequency plane, compute it
needed = self.fft_size*(1. - self.overlap)
realizable = int(floor(available/needed))
if realizable > 0:
spn = zeros((len(self.freq), realizable), dtype=float64)
for i in range(realizable):
floatdata = self.audiobuffer.data_indexed(self.old_index, self.fft_size)
# for now, take the first channel only
floatdata = floatdata[0,:]
# FIXME We should allow here for more intelligent transforms, especially when the log freq scale is selected
spn[:, i] = self.proc.analyzelive(floatdata)
self.old_index += int(needed)
w = tile(self.w, (1, realizable))
norm_spectrogram = self.scale_spectrogram(self.log_spectrogram(spn) + w)
self.PlotZoneImage.addData(self.freq, norm_spectrogram)
self.PlotZoneImage.updatePlot()
# thickness of a frequency column depends on FFT size and window overlap
# hamming window with 75% overlap provides good quality (Perfect reconstruction,
# aliasing from side lobes only, 42 dB channel isolation)
# number of frequency columns that we keep depends on the time history that the user has chosen
# actual displayed spectrogram is a scaled version of the time-frequency plane
def setminfreq(self, freq):
self.minfreq = freq
self.PlotZoneImage.setfreqrange(self.minfreq, self.maxfreq)
def setmaxfreq(self, freq):
self.maxfreq = freq
self.PlotZoneImage.setfreqrange(self.minfreq, self.maxfreq)
self.proc.set_maxfreq(freq)
self.update_weighting()
self.freq = self.proc.get_freq_scale()
def setfftsize(self, fft_size):
self.fft_size = fft_size
self.proc.set_fftsize(fft_size)
self.update_weighting()
self.freq = self.proc.get_freq_scale()
self.dT_s = self.fft_size*(1. - self.overlap)/float(SAMPLING_RATE)
self.PlotZoneImage.settimerange(self.timerange_s, self.dT_s)
sfft_rate_frac = Fraction(SAMPLING_RATE, self.fft_size)/(Fraction(1) - self.overlap_frac)/1000
self.PlotZoneImage.set_sfft_rate(sfft_rate_frac)
def setmin(self, value):
self.spec_min = value
self.PlotZoneImage.setspecrange(self.spec_min, self.spec_max)
def setmax(self, value):
self.spec_max = value
self.PlotZoneImage.setspecrange(self.spec_min, self.spec_max)
def setweighting(self, weighting):
self.weighting = weighting
self.PlotZoneImage.setweighting(weighting)
self.update_weighting()
def update_weighting(self):
A, B, C = self.proc.get_freq_weighting()
if self.weighting is 0:
self.w = 0.
elif self.weighting is 1:
self.w = A
elif self.weighting is 2:
self.w = B
else:
self.w = C
self.w.shape = (len(self.w), 1)
def settings_called(self, checked):
self.settings_dialog.show()
def saveState(self, settings):
self.settings_dialog.saveState(settings)
def restoreState(self, settings):
self.settings_dialog.restoreState(settings)
# slot
def timerangechanged(self, value):
self.timerange_s = value
self.PlotZoneImage.settimerange(self.timerange_s, self.dT_s)
# slot
def canvasWidthChanged(self, width):
self.canvas_width = width
class Spectrogram_Widget(QtWidgets.QWidget):
def __init__(self, parent):
super().__init__(parent)
self.setObjectName("Spectrogram_Widget")
self.gridLayout = QtWidgets.QGridLayout(self)
self.gridLayout.setObjectName("gridLayout")
self.PlotZoneImage = ImagePlot(self)
self.PlotZoneImage.setObjectName("PlotZoneImage")
self.gridLayout.addWidget(self.PlotZoneImage, 0, 1, 1, 1)
self.audiobuffer = None
# initialize the class instance that will do the fft
self.proc = audioproc()
self.maxfreq = DEFAULT_MAXFREQ
self.proc.set_maxfreq(self.maxfreq)
self.minfreq = DEFAULT_MINFREQ
self.fft_size = 2**DEFAULT_FFT_SIZE * 32
self.proc.set_fftsize(self.fft_size)
self.spec_min = DEFAULT_SPEC_MIN
self.spec_max = DEFAULT_SPEC_MAX
self.weighting = DEFAULT_WEIGHTING
self.update_weighting()
self.freq = self.proc.get_freq_scale()
self.timerange_s = DEFAULT_TIMERANGE
self.canvas_width = 100.
self.old_index = 0
self.overlap = 3. / 4.
self.overlap_frac = Fraction(3, 4)
self.dT_s = self.fft_size * (1. - self.overlap) / float(SAMPLING_RATE)
self.PlotZoneImage.setlog10freqscale() # DEFAULT_FREQ_SCALE = 1 #log10
self.PlotZoneImage.setfreqrange(self.minfreq, self.maxfreq)
self.PlotZoneImage.setspecrange(self.spec_min, self.spec_max)
self.PlotZoneImage.setweighting(self.weighting)
self.PlotZoneImage.settimerange(self.timerange_s, self.dT_s)
self.update_jitter()
sfft_rate_frac = Fraction(SAMPLING_RATE, self.fft_size) / (
Fraction(1) - self.overlap_frac) / 1000
self.PlotZoneImage.set_sfft_rate(sfft_rate_frac)
# initialize the settings dialog
self.settings_dialog = Spectrogram_Settings_Dialog(self)
AudioBackend().underflow.connect(
self.PlotZoneImage.plotImage.canvasscaledspectrogram.syncOffsets)
self.last_data_time = 0.
self.mustRestart = False
# method
def set_buffer(self, buffer):
self.audiobuffer = buffer
self.old_index = self.audiobuffer.ringbuffer.offset
def log_spectrogram(self, sp):
# Note: implementing the log10 of the array in Cython did not bring
# any speedup.
# Idea: Instead of computing the log of the data, I could pre-compute
# a list of values associated with the colormap, and then do a search...
epsilon = 1e-30
return 10. * log10(sp + epsilon)
# scale the db spectrum from [- spec_range db ... 0 db] to [0..1] (do not clip, will be down after resampling)
def scale_spectrogram(self, sp):
return (sp - self.spec_min) / (self.spec_max - self.spec_min)
def handle_new_data(self, floatdata):
# we need to maintain an index of where we are in the buffer
index = self.audiobuffer.ringbuffer.offset
self.last_data_time = self.audiobuffer.lastDataTime
available = index - self.old_index
if available < 0:
# ringbuffer must have grown or something...
available = 0
self.old_index = index
# if we have enough data to add a frequency column in the time-frequency plane, compute it
needed = self.fft_size * (1. - self.overlap)
realizable = int(floor(available / needed))
if realizable > 0:
spn = zeros((len(self.freq), realizable), dtype=float64)
for i in range(realizable):
floatdata = self.audiobuffer.data_indexed(
self.old_index, self.fft_size)
# for now, take the first channel only
floatdata = floatdata[0, :]
# FFT transform
spn[:, i] = self.proc.analyzelive(floatdata)
self.old_index += int(needed)
w = tile(self.w, (1, realizable))
norm_spectrogram = self.scale_spectrogram(
self.log_spectrogram(spn) + w)
self.PlotZoneImage.addData(self.freq, norm_spectrogram,
self.last_data_time)
if self.mustRestart:
self.PlotZoneImage.restart()
self.mustRestart = False
# thickness of a frequency column depends on FFT size and window overlap
# hamming window with 75% overlap provides good quality (Perfect reconstruction,
# aliasing from side lobes only, 42 dB channel isolation)
# number of frequency columns that we keep depends on the time history that the user has chosen
# actual displayed spectrogram is a scaled version of the time-frequency plane
def canvasUpdate(self):
if not self.isVisible():
return
self.PlotZoneImage.draw()
def update_jitter(self):
audio_jitter = 2 * float(FRAMES_PER_BUFFER) / SAMPLING_RATE
analysis_jitter = self.fft_size * (1. - self.overlap) / SAMPLING_RATE
canvas_jitter = audio_jitter + analysis_jitter
# print audio_jitter, analysis_jitter, canvas_jitter
self.PlotZoneImage.plotImage.set_jitter(canvas_jitter)
def pause(self):
self.PlotZoneImage.pause()
def restart(self):
# defer the restart until we get data from the audio source (so that a fresh lastdatatime is passed to the spectrogram image)
self.mustRestart = True
def setminfreq(self, freq):
self.minfreq = freq
self.PlotZoneImage.setfreqrange(self.minfreq, self.maxfreq)
def setmaxfreq(self, freq):
self.maxfreq = freq
self.PlotZoneImage.setfreqrange(self.minfreq, self.maxfreq)
self.proc.set_maxfreq(freq)
self.update_weighting()
self.freq = self.proc.get_freq_scale()
def setfftsize(self, fft_size):
self.fft_size = fft_size
self.proc.set_fftsize(fft_size)
self.update_weighting()
self.freq = self.proc.get_freq_scale()
self.dT_s = self.fft_size * (1. - self.overlap) / float(SAMPLING_RATE)
self.PlotZoneImage.settimerange(self.timerange_s, self.dT_s)
sfft_rate_frac = Fraction(SAMPLING_RATE, self.fft_size) / (
Fraction(1) - self.overlap_frac) / 1000
self.PlotZoneImage.set_sfft_rate(sfft_rate_frac)
self.update_jitter()
def setmin(self, value):
self.spec_min = value
self.PlotZoneImage.setspecrange(self.spec_min, self.spec_max)
def setmax(self, value):
self.spec_max = value
self.PlotZoneImage.setspecrange(self.spec_min, self.spec_max)
def setweighting(self, weighting):
self.weighting = weighting
self.PlotZoneImage.setweighting(weighting)
self.update_weighting()
def update_weighting(self):
A, B, C = self.proc.get_freq_weighting()
if self.weighting is 0:
self.w = array([0.])
elif self.weighting is 1:
self.w = A
elif self.weighting is 2:
self.w = B
else:
self.w = C
self.w.shape = (len(self.w), 1)
def settings_called(self, checked):
self.settings_dialog.show()
def saveState(self, settings):
self.settings_dialog.saveState(settings)
def restoreState(self, settings):
self.settings_dialog.restoreState(settings)
# slot
def timerangechanged(self, value):
self.timerange_s = value
self.PlotZoneImage.settimerange(self.timerange_s, self.dT_s)
# slot
def canvasWidthChanged(self, width):
self.canvas_width = width
