def output2_plot(self):
if 'exportpath' in globals() and os.path.exists(exportpath):
spectrogram(os.path.join(exportpath, 'SEN_2.wav'),
'Output Audio Sample Two', 'output02.png')
Pixmap = QPixmap('output02.png')
self.Spectrogram.setPixmap(Pixmap)
else:
QMessageBox.about(self, 'Warning',
"First run Acoustic Separation Model")
def input_plot(self):
if 'filename' in globals() and os.path.exists(filename):
spectrogram(os.path.join(filename), 'Input Audio Sample',
'input.png')
Pixmap = QPixmap('input.png')
self.Spectrogram.setPixmap(Pixmap)
else:
QMessageBox.about(
self, 'Warning',
"First add external audio file or record audio file")
def plotAll(m):
pitch = []
end = 0
for index, r in enumerate(m.speechSegment):
pitch = pitch + [0] * (r[0] - end) + m.pitch[index]
end = r[1]
pitch = pitch + [0] * (m.frameNum - m.speechSegment[-1][1])
pl.subplot(611)
pl.plot(m.absVolume)
#pl.xlabel('Frame Num')
pl.ylabel('Volume')
#pl.grid(True)
for s in m.speechSegment:
pl.plot([s[0], s[0]], [0, max(m.absVolume)], color='red')
pl.plot([s[1], s[1]], [0, max(m.absVolume)], color='green')
pl.subplot(612)
pl.plot(m.zcr)
pl.ylabel('Zero Cross Rate')
#pl.xlabel('Frame Num')
pl.subplot(613)
pl.plot(pitch)
#pl.xlabel('Frame Num')
pl.ylabel('Pitch')
for index, s in enumerate(m.pitchSeg):
for p in s:
pl.plot([
p[0] + m.speechSegment[index][0],
p[0] + m.speechSegment[index][0]
], [0, 500],
color='red')
pl.plot([
p[1] + m.speechSegment[index][0],
p[1] + m.speechSegment[index][0]
], [0, 500],
color='green')
pl.subplot(614)
pl.plot(m.f1)
pl.ylabel('Formant 1')
pl.subplot(615)
spectrogram(m)
pl.ylabel("spectrogram")
pl.xlabel('Frame Num')
pl.subplot(616)
pl.ylabel("Energy Below 250Hz")
pl.plot(m.energyBelow250)
pl.show()
def spectrogram_process(self):
"""Processing spectrogram
:returns: TODO
"""
spectrogram_obj = spectrogram.spectrogram(self._aiff_datas, self._temporal_data_point_list,
self._temporal_point)
return np.asarray(spectrogram_obj.spectrogram_generate())
def calculateSpectro(sound,
spec_sample_rate=1000,
freq_spacing=210,
min_freq=0,
max_freq=10000):
t,f,spec,spec_rms = spectrogram(sound, 44100, spec_sample_rate=spec_sample_rate, freq_spacing=freq_spacing, \
min_freq=min_freq, max_freq=max_freq, cmplx=True)
spectro = 20 * np.log10(np.abs(spec))
return spectro
def plotAll(m):
pitch = []
end = 0
for index,r in enumerate(m.speechSegment):
pitch = pitch + [0]*(r[0]-end) + m.pitch[index]
end = r[1]
pitch = pitch + [0]*(m.frameNum-m.speechSegment[-1][1])
pl.subplot(611)
pl.plot(m.absVolume)
#pl.xlabel('Frame Num')
pl.ylabel('Volume')
#pl.grid(True)
for s in m.speechSegment:
pl.plot([s[0],s[0]],[0,max(m.absVolume)],color='red')
pl.plot([s[1],s[1]],[0,max(m.absVolume)],color='green')
pl.subplot(612)
pl.plot(m.zcr)
pl.ylabel('Zero Cross Rate')
#pl.xlabel('Frame Num')
pl.subplot(613)
pl.plot(pitch)
#pl.xlabel('Frame Num')
pl.ylabel('Pitch')
for index,s in enumerate(m.pitchSeg):
for p in s:
pl.plot([p[0]+m.speechSegment[index][0],p[0]+m.speechSegment[index][0]],[0,500],color='red')
pl.plot([p[1]+m.speechSegment[index][0],p[1]+m.speechSegment[index][0]],[0,500],color='green')
pl.subplot(614)
pl.plot(m.f1)
pl.ylabel('Formant 1')
pl.subplot(615)
spectrogram(m)
pl.ylabel("spectrogram")
pl.xlabel('Frame Num')
pl.subplot(616)
pl.ylabel("Energy Below 250Hz")
pl.plot(m.energyBelow250)
pl.show()
def _audio_file_spectrogram(self, _file, nfft, duration, overlap):
# first get the spec_params and let the client setup the canvas
fs = _file.sample_rate
spec_params = get_audio_spectrogram_params(_file, fs, duration, nfft,
overlap)
self.send_spectrogram_new(spec_params)
# now lets compute the spectrogram and send it over
data = _file[:spec_params.nsamples].sum(axis=1)
for chunk in grouper(data, spec_params.chunksize, spec_params.nfft):
chunk = np.array(chunk)
spec = spectrogram(chunk, spec_params)
self.send_spectrogram_update(spec)
def _audio_file_spectrogram(self, _file, nfft, duration, overlap):
# first get the spec_params and let the client setup the canvas
fs = _file.sample_rate
spec_params = get_audio_spectrogram_params(
_file, fs, duration, nfft, overlap)
self.send_spectrogram_new(spec_params)
# now lets compute the spectrogram and send it over
data = _file[:spec_params.nsamples].sum(axis=1)
for chunk in grouper(data, spec_params.chunksize, spec_params.nfft):
chunk = np.array(chunk)
spec = spectrogram(chunk, spec_params)
self.send_spectrogram_update(spec)
def __process_wav(self, wav, label_dict, class_name, f_label, f_label_bg, f_meta, additional_meta=""):
print("Preparing file: ", wav)
y, sr = librosa.load(wav)
spec = spectrogram(y)
spec, bg = self.bg_subtraction(spec)
chunks = self.make_chunks(spec)
name, ftype = wav.split(".")
name = name.split("/")[-1]
for i, chunk in enumerate(chunks):
out_fname = str.format("{}/{}_{}.pkl", self.out_path, name, i)
pickle.dump(chunk, open(out_fname, "wb"), protocol=2)
self.label_lock.acquire()
f_label.write(out_fname + " " + str(label_dict[class_name]) + "\n")
f_meta.write("{} {} {} {} {}\n".format(out_fname, sr, chunk.shape[0],
chunk.shape[1], additional_meta.encode("utf-8")))
f_label.flush()
f_meta.flush()
self.label_lock.release()
if bg.shape[1] == 0 or spec.shape[1] == 0:
return
bg = np.true_divide(bg, np.max(spec))
chunks_bg = self.make_chunks(bg)
for i, chunk in enumerate(chunks_bg):
out_fname = str.format("{}/bg/{}_bg_{}.pkl", self.out_path, name, i)
self.bg_lock.acquire()
f_label_bg.write(out_fname + " " + str(label_dict[class_name]) + "\n")
f_label_bg.flush()
self.bg_lock.release()
pickle.dump(chunk, open(out_fname, "wb"), protocol=2)
# Let's look at some difference spectrograms...
voiced_events = []
unvoiced_events = []
trial_events = grp.events().query_events( 'TRIAL' )
for event in trial_events:
if 'DAH' in event['name']: voiced_events.append( event )
if 'GAH' in event['name']: voiced_events.append( event )
if 'BAH' in event['name']: voiced_events.append( event )
if 'TAH' in event['name']: unvoiced_events.append( event )
if 'KAH' in event['name']: unvoiced_events.append( event )
if 'PAH' in event['name']: unvoiced_events.append( event )
voiced_events = np.array( voiced_events )
unvoiced_events = np.array( unvoiced_events )
from spectrogram import spectrogram, _plot_specgram
voiced_spec, f, t = spectrogram( grp.eeg(), 'AMIC15', voiced_events, progress_bar = True )
unvoiced_spec, f, t = spectrogram( grp.eeg(), 'AMIC15', unvoiced_events, progress_bar = True )
diff_spec = np.subtract( voiced_spec, unvoiced_spec )
_plot_specgram( voiced_spec, f, t, title = "Voiced",
low_freq = 0, high_freq = 2000, clamp = 3.0 )
_plot_specgram( unvoiced_spec, f, t, title = "Unvoiced",
low_freq = 0, high_freq = 2000, clamp = 3.0 )
_plot_specgram( diff_spec, f, t, title = "Voiced - Unvoiced",
low_freq = 0, high_freq = 2000, clamp = 3.0 )
# Sonification!
from util import listen
listen( grp.eeg()[ grp.events().query('KAH', length = 10000), 'AMIC15' ], grp.eeg().get_rate() )
listen( grp.eeg()[ grp.events().query('TRIAL', length = 10000), 'PMIC9' ], grp.eeg().get_rate() )
Segment / np.abs(Segment).max() * np.max(allSegments)
for Segment in allSegments
])
allSpectros = np.array(
[np.ravel(calculateSpectro(Seg)) for Seg in allSegments])
allSpectros = allSpectros - allSpectros.max()
allFund = np.load('fund_Len500.npy')
nonanIndFund = ~np.isnan(allFund)
#lessInd = np.where(allFund<800)[0]
nonanInd = [d and m for d, m in zip(nonanIndFund, nonanIndFund)]
#allFundFiltered = allFund[nonanInd]
##print(allFundFiltered[:100])
birdName = birdName[nonanInd]
allSpectros = allSpectros[nonanInd]
(tDebug, freqDebug, specDebug, rms) = spectrogram(allPeriods[0],
44100,
1000.0,
210,
min_freq=0,
max_freq=10000,
nstd=2,
cmplx=True)
plot_spectrogram(tDebug, freqDebug, specDebug)
plt.show()
Ind = np.arange(len(birdName))
np.random.shuffle(Ind)
birdNameShuffled = birdName[Ind]
allSpectros = allSpectros[Ind]
#scores, pValues, nhighScores, ncounter = pwLogiRegression(allSpectros, birdNameShuffled, nonanIndFeature=None, cv=True, printijScores=False)
import os
import numpy as np
from functions import list_songs
from spectrogram import spectrogram
os.makedirs('./spectro', exist_ok=True)
diff = 'expert'
precision = 2
for f in list_songs(diff):
H = spectrogram(f, precision)
np.save(f'./spectro/{f}', H)
def load_npy(self,np_path):
#print("Loading npy file: ", np_path)
y = pickle.load(open(np_path, "rb"))
return spectrogram(y)
chirp = Signal('chirp', 'LFM_1K_5K.wav')
chirp.generateValsFromFile()
chirp.values /= np.max(chirp.values)
chirp.values -= np.mean(chirp.values)
outside = Signal('outside', '171013104635_1_B.wav')
outside.generateValsFromFile()
outside.values /= np.max(outside.values)
plt.figure(figsize=(17 * 2 / 3, 22 * 2 / 3))
plt.tight_layout()
# plot signals' spectrograms
plt.subplot(321)
plt.title("Chirp")
specs, f, t = spectrogram(chirp, 1000, 200)
specs = specs[:, ::20]
t = t[::20]
print('Heatmap size:', np.shape(specs))
t, f = np.meshgrid(t, f)
plt.xlabel('Time (s)')
plt.ylabel('Frequency (Hz)')
plt.ylim(0, 6000)
plt.xlim(0, 0.25)
plt.pcolormesh(t, f, specs, vmin=-150, cmap="BuPu")
print("@debug created first spectrogram")
plt.subplot(322)
plt.title("Outside")
specs2, f2, t2 = spectrogram(outside, 1000, 200)
specs2 = specs2[:, ::20]
