def classifyFile(dirs, soundFile, windowFile):
print '### Classifying:', soundFile
winsize = windowFile['windows'][0]['size']
winstride = windowFile['windows'][0]['stride']
# Preprocess the file
prepFile(soundFile)
# Read the file
filepointer = h5py.File(soundFile, 'r')
signals = files.signalsFromHDF5(soundFile)
startEndPairs = makeStartEndPairs(winsize, winstride, signals)
windowDic = makeWindowDic(startEndPairs, signals, winsize)
windowPredictions = classifyWindows(dirs, startEndPairs, signals,
windowDic, winsize)
filePredictions = {}
for key in windowPredictions.keys():
filePredictions[key] = sum(windowPredictions[key]) / len(
windowPredictions[key])
windowPredictions['windows'] = startEndPairs
return windowPredictions, filePredictions
def classifyFile(dirs, soundFile, windowFile):
print '### Classifying:', soundFile
winsize = windowFile['windows'][0]['size']
winstride = windowFile['windows'][0]['stride']
# Preprocess the file
prepFile(soundFile)
# Read the file
filepointer = h5py.File(soundFile, 'r')
signals = files.signalsFromHDF5(soundFile)
startEndPairs = makeStartEndPairs(winsize, winstride, signals)
windowDic = makeWindowDic(startEndPairs, signals, winsize)
windowPredictions = classifyWindows(dirs, startEndPairs, signals, windowDic, winsize)
filePredictions = {}
for key in windowPredictions.keys():
filePredictions[key] = sum(windowPredictions[key])/len(windowPredictions[key])
windowPredictions['windows'] = startEndPairs
return windowPredictions, filePredictions
def prepFile(f):
print "file: " + f
if os.path.isfile(f):
# Read information from the hdf5
signals = files.signalsFromHDF5(f)
# Original signal
original = signals['energy'][:, :]
# Get the zero/infinity columns out
original = np.transpose(original)
zeroColLeft = []
zeroColRight = []
while max(original[0]) <= 0:
zeroColLeft.append(original[0])
original = original[1:]
while max(original[-1]) <= 0:
zeroColRight.append(original[-1])
original = original[:-2]
original = np.transpose(original)
# preprocess signal
morphology = p.dilate(p.erode(original))
# Put the zeros/infinities back
if len(zeroColLeft) > 0 or len(zeroColRight) > 0:
if len(zeroColLeft) > 0:
morphology = np.column_stack(
(np.transpose(zeroColLeft), morphology))
if len(zeroColRight) > 0:
morphology = np.column_stack(
(morphology, np.transpose(zeroColRight)))
# Store the preprocessed signal
if 'morphology' in signals:
signals['morphology'] = morphology
else:
with h5py.File(f, 'r+') as filepointer:
filepointer.create_dataset('morphology', data=morphology)
# Foreground background
definitions = [
bgconfig.getDefaults(tau) for tau in bgconfig.tau(0.5, 4)
]
bgmodels = [bgmodel.BGModel(
d['tau'],
d,
) for d in definitions]
for model in bgmodels:
response = model.calculate(original)
# Put the zeros/infinities back
if len(zeroColLeft) > 0 or len(zeroColRight) > 0:
if len(zeroColLeft) > 0:
response = np.column_stack(
(np.transpose(zeroColLeft), response))
if len(zeroColRight) > 0:
response = np.column_stack(
(response, np.transpose(zeroColRight)))
# np.save(prepDir + 'tau {}'.format(model.name), response)
if 'tau {}'.format(model.name) in signals:
signals['tau {}'.format(model.name)] = response
else:
with h5py.File(f, 'r+') as filepointer:
filepointer.create_dataset('tau {}'.format(model.name),
data=response)
label = 0
images = []
labels = []
for subdir in dirs:
d = os.path.join(root, subdir)
hdf5dir = os.path.join(d, 'ptne/hdf5/2016-06-02')
h5files = [f for f in os.listdir(hdf5dir) if '.wav.3' in f]
if 'Pouring' in hdf5dir or 'Church bells' in hdf5dir or 'Glass' in hdf5dir:
print 'skipping ', hdf5dir
continue
print 'Working on ', hdf5dir
for h5file in h5files:
signals = files.signalsFromHDF5(os.path.join(hdf5dir, h5file))
energy = signals['energy']
pulse = signals['tfmax']
tone = signals['tsmax']
max = np.max([tone, energy, pulse])
min = np.min([tone, energy, pulse])
full_im = (np.stack([tone, pulse, energy], axis=0).transpose(
(1, 2, 0)) - min) / (max - min)
full_im = imresize(full_im, (109, 115))
images.append(full_im)
labels.append(label)
#plt.imshow(255 - full_im)
#plt.pause(0.05)
if len(sys.argv) != 2:
print "Usage: python processSoundFile.py <filename.hdf5>"
with open('labeling/windows.yaml', 'r') as f:
windows = yaml.load(f)
######### DIRECTORIES #########
dirs = {}
dirs['networks'] = 'classifiers/trained_nets/'
dirs['fihs'] = 'classifiers/fih'
######### READING FILE #########
soundFile = sys.argv[1]
filepointer = h5py.File(soundFile, 'r+')
signals = files.signalsFromHDF5(soundFile)
######### CLASSIFICATION #########
windowPredictions, filePredictions = classifyFile(dirs, soundFile, windows)
# windowPredictions = pickle.load( open( "windowPredictions.pickle", "rb" ) )
sudden = True
loudness = True
######### Loudness and sudden features #########
hist = makeWavelet(signals)
if sudden:
differenceFactors, loudnessFactors = getSuddenSounds(hist, signals)
for index, window in enumerate(windowPredictions['windows']):
diffFactor = float(sum(differenceFactors[window[0]:window[1]])) / float(window[1]-window[0])
import sys, glob, shutil
import h5py, numpy as np
import os, matplotlib.pyplot as plt
# Change search directory for modules to include pycpsp in parent
sys.path.insert(1, os.path.join(sys.path[0], '..'))
import pycpsp.files as files
import pycpsp.bgmodel.config as bgconfig
import pycpsp.bgmodel.bgmodel as bgmodel
if __name__ == '__main__':
f = sys.argv[1]
if os.path.isfile(f):
# Read information from the hdf5
signals = files.signalsFromHDF5(f)
# Original signal
original = signals['energy'][:, :]
print original.shape
# Get the zero/infinity columns out
original = np.transpose(original)
zeroColLeft = []
zeroColRight = []
while max(original[0]) <= 0:
zeroColLeft.append(original[0])
original = original[1:]
while max(original[-1]) <= 0:
zeroColRight.append(original[-1])
original = original[:-2]
original = np.transpose(original)
print original.shape
for model in bgmodels:
response = model.calculate(refsignal)
savePlot(response, 'tau {}'.format(model.name), filename, metadata, run)
# plot.plot1D('frequency-averaged energy', np.sum(refsignal, axis=0) / refsignal.shape[0], ylim=[0,60])
# plot.plot1D('frequency-averaged bg model tau:{}'.format(model.name), np.sum(response, axis=0) / refsignal.shape[0], ylim=[0,60])
savePlot(refbg - response, 'tau 0.1 minus tau {}'.format(model.name), filename, metadata, run)#, mask=[0,20], scale=None)
if __name__ == '__main__':
if len(sys.argv) < 1:
print("usage:", sys.argv[0], "<name.hdf5>")
sys.exit(0)
filename = sys.argv[1]
filepointer = h5py.File(filename, 'r')
signals = files.signalsFromHDF5(filename)
metadata = files.metadataFromHDF5(filename)
# Original image
original = signals['energy'][:, metadata.get('starttime 0'):]
refsignal = plot.imgmask(plot.imscale(signals['energy'],[0,60]), [20,80])
# preprocess images
eroded = erode(refsignal)
dilated = dilate(refsignal)
combined = dilate(erode(refsignal))
foregroundBackground(refsignal, f)
# Save the plots
savePlot(refsignal, 'refsignal', metadata)
savePlot(original, 'original', metadata)
label = 0
images = []
labels = []
for subdir in dirs:
d = os.path.join(root, subdir)
hdf5dir = os.path.join(d, 'ptne/hdf5/2016-06-02')
h5files = [f for f in os.listdir(hdf5dir) if '.wav.3' in f]
if 'Pouring' in hdf5dir or 'Church bells' in hdf5dir or 'Glass' in hdf5dir:
print 'skipping ', hdf5dir
continue
print 'Working on ', hdf5dir
for h5file in h5files:
signals = files.signalsFromHDF5(os.path.join(hdf5dir, h5file))
energy = signals['energy']
pulse = signals['tfmax']
tone = signals['tsmax']
max = np.max([tone, energy, pulse])
min = np.min([tone, energy, pulse])
full_im = (np.stack([tone, pulse, energy], axis=0).transpose((1, 2, 0)) - min) / (max - min)
full_im = imresize(full_im, (109, 115))
images.append(full_im)
labels.append(label)
#plt.imshow(255 - full_im)
#plt.pause(0.05)
label += 1
if __name__ == '__main__':
if len(sys.argv) != 2:
print "Usage: python processSoundFile.py <filename.hdf5>"
with open('labeling/windows.yaml', 'r') as f:
windows = yaml.load(f)
######### DIRECTORIES #########
dirs = {}
dirs['networks'] = 'classifiers/trained_nets/'
dirs['fihs'] = 'classifiers/fih'
######### READING FILE #########
soundFile = sys.argv[1]
filepointer = h5py.File(soundFile, 'r+')
signals = files.signalsFromHDF5(soundFile)
######### CLASSIFICATION #########
windowPredictions, filePredictions = classifyFile(dirs, soundFile, windows)
# windowPredictions = pickle.load( open( "windowPredictions.pickle", "rb" ) )
sudden = True
loudness = True
######### Loudness and sudden features #########
hist = makeWavelet(signals)
if sudden:
differenceFactors, loudnessFactors = getSuddenSounds(hist, signals)
for index, window in enumerate(windowPredictions['windows']):
diffFactor = float(sum(
differenceFactors[window[0]:window[1]])) / float(window[1] -