def extractFeatures(input_signal):
"""extract features from the cleaned signal.
:param cleaned signal
:return: features list"""
# compute mfcc list
if len(input_signal) == 0:
print("cleaned signal is empty")
return input_signal
mfcc_list = np.array(mfcc(input_signal, samplerate=prm.params["sample_rate"].get(), winlen=0.032, winstep=0.016, numcep=30,
nfilt=55, nfft=2048, lowfreq=0, highfreq=6000, preemph=0.95, ceplifter=22, appendEnergy=True ))
extractor = LPCExtractor(prm.params["sample_rate"].get(), 32, 16, 30, 0.95)
lpcc = extractor.extract(input_signal)
pitch = extract_pitch(input_signal)
# Cepstral Mean Normalization @TODO: WHY IS THIS NOT HELPING??
if 0:
mean_mfcc = np.mean(mfcc_list.T, 1)
std_mfcc = np.std(mfcc_list.T, 1)
for i in range(len(mfcc_list)):
for j in range(len(mfcc_list[i])):
mfcc_list[i][j] = (mfcc_list[i][j]-mean_mfcc[j])/std_mfcc[j]
# print np.shape(mfcc_list[i-1]), np.shape(mean_mfcc), np.shape(std_mfcc)
N = 2
delta_list = delta(mfcc_list, N)
ddelta_list = delta(delta_list, N)
# do not keep first coeff (energy)
features_list = list()
for k in range(len(mfcc_list)):
# features_list += [np.hstack((mfcc_list[k][0:], lpcc[k][0:]))]
features_list += [mfcc_list[k][0:]]
# features_list += [lpcc[k][0:]]
# features_list += [np.hstack((mfcc_list[k][0:], delta_list[k][0:], ddelta_list[k][0:]))]
# print np.shape(mfcc_list), np.shape(features_list)
# dont return nan
# @TODO WHY DOES THIS HAPPEN?
for row in features_list:
for cell in row:
if cell != cell:
print "Cell is nan (see feature extraction):", str(cell)
return []
full_features_list = []
full_features_list = list(np.ravel(features_list))
full_features_list.extend([pitch]*30) # do we need to append this multiple times to ensure that the forest selects it?
return full_features_list
def comp_feat(self, sig, rate):
'''
compute the features
Args:
sig: the audio signal as a 1-D numpy array
rate: the sampling rate
Returns:
the features as a [seq_length x feature_dim] numpy array
'''
#snip the edges
sig = snip(sig, rate, float(self.conf['winlen']),
float(self.conf['winstep']))
feat, energy = base.mfcc(sig, rate, self.conf)
if self.conf['include_energy'] == 'True':
feat = np.append(feat, energy[:, np.newaxis], 1)
if self.conf['dynamic'] == 'delta':
feat = base.delta(feat)
elif self.conf['dynamic'] == 'ddelta':
feat = base.ddelta(feat)
elif self.conf['dynamic'] != 'nodelta':
raise Exception('unknown dynamic type')
return feat
def comp_feat(self, sig, rate):
'''
compute the features
Args:
sig: the audio signal as a 1-D numpy array
rate: the sampling rate
Returns:
the features as a [seq_length x feature_dim] numpy array
'''
feat, energy = base.logfbank(sig, rate, self.conf)
if self.conf['include_energy'] == 'True':
feat = np.append(feat, energy[:, np.newaxis], 1)
if self.conf['dynamic'] == 'delta':
feat = base.delta(feat)
elif self.conf['dynamic'] == 'ddelta':
feat = base.ddelta(feat)
elif self.conf['dynamic'] != 'nodelta':
raise Exception('unknown dynamic type')
#mean and variance normalize the features
if self.conf['mvn'] == 'True':
feat = (feat - feat.mean(0)) / feat.std(0)
return feat
def main():
auth = login(url, user, password)
host_group = {}
cc = 0
if auth:
file = open(path,'w')
file.write(','.join(['Severity','Last change','Age','Host','Name','History Alerts Num','Groups','Templates'])+'\n')
params = {
#"output": ["triggerid", "status", "type","value", "description", "templateid"],
"output": "extend",
"only_true": True, #Return only triggers that have recently been in a problem state.
"monitored": True, #Return only enabled triggers that belong to monitored hosts and contain only enabled items.
"selectItems": "extend",
"selectHosts": ["hostid","status","name"],
"selectGroups":["name"],
"selectLastEvent": "extend",
#"filter": {
# "trigger": trigger_list,
# }
}
triggers = trigger_get(url, auth, params)
triggers.sort(key=lambda x:x['lastchange'], reverse=True)
for trigger in triggers:
if trigger['value'] == '1' and trigger['status'] == '0':
cc += 1
#print trigger
#itemids = [item['itemid'] for item in trigger['items']]
#print itemids
'''
sub_triggers = trigger_get(auth, {
"output": "extend",
"itemids" : itemids,
})
if len(sub_triggers) > 1:
print sub_triggers
'''
events = event_get(url, auth, {
"output": "extend",
"triggerids" : [trigger['triggerid']],
"value": "1",
})
templates = template_get(url, auth, {
"output": "extend",
"hostids": trigger['hosts'][0]['hostid'],
})
print trigger['hosts'][0]['name'], datetime.now() - trigger['lastchange'], delta(datetime.now() - trigger['lastchange'])
res = [trigger['priority'].capitalize(), trigger['lastchange'], delta(datetime.now() - trigger['lastchange']), trigger['hosts'][0]['name'], trigger['description'].replace('{HOST.NAME}',trigger['hosts'][0]['name']), len(events)-1, '/'.join([g['name'] for g in trigger['groups']]), '/'.join([t['name'] for t in templates])]
line = ','.join([not isinstance(r,str) and str(r) or r for r in res])
file.write(line.decode("utf-8").encode('GBK')+'\n')
file.flush()
file.close()
#print(cc)
return cc
def comp_feat(self, sig, rate):
'''
compute the features
Args:
sig: the audio signal as a 1-D numpy array
rate: the sampling rate
Returns:
the features as a [seq_length x feature_dim] numpy array
'''
feat, energy = base.mfcc(sig, rate, self.conf)
# write the wav to temporary location and invoke external pitch extractor.
# make sure 'reaper' is in your $PATH
tempdir = os.path.join('/tmp', str(os.getpid()))
if not os.path.isdir(tempdir):
os.makedirs(tempdir)
name = 'mix'
wav.write(os.path.join(tempdir, name + '.wav'), rate, np.int16(sig))
os.system('reaper -i ' + os.path.join(tempdir, name + '.wav') +
' -f ' + os.path.join(tempdir, name + '.txt -a -e 0.01'))
pitch = np.loadtxt(os.path.join(tempdir, name + '.txt'), skiprows=7)[:,
2]
pitch = np.pad(pitch, (0, max(0, feat.shape[0] - pitch.shape[0])),
'edge')
# linear interpolation in voiceless regions
voiceless = np.where(pitch == -1)[0]
jump = np.where((voiceless[1:] - voiceless[:-1]) > 1)[0]
segments = np.split(voiceless, jump + 1)
for seg in segments:
idx1 = seg[0] - 1
idx2 = seg[-1] + 1
val1 = -1
val2 = -1
if idx1 >= 0:
val1 = pitch[idx1]
if idx2 < pitch.size:
val2 = pitch[idx2]
if val1 == -1: #segment starts at utterence start
val1 = val2
if val2 == -1: # segment ends at utterance end
val2 = val1
if val1 == -1: #segment is the whole utterance => make up a value
val1 = 150
val2 = 150
#interpolate
pitch[seg] = (val2 - val1) * (np.array(seg) -
idx1) / float(idx2 - idx1) + val1
feat = np.append(feat, pitch[:feat.shape[0], np.newaxis], 1)
if self.conf['include_energy'] == 'True':
feat = np.append(feat, energy[:, np.newaxis], 1)
if self.conf['dynamic'] == 'delta':
feat = base.delta(feat)
elif self.conf['dynamic'] == 'ddelta':
feat = base.ddelta(feat)
elif self.conf['dynamic'] != 'nodelta':
raise Exception('unknown dynamic type')
#mean and variance normalize the features
if self.conf['mvn'] == 'True':
feat = (feat - feat.mean(0)) / (
feat.std(0) + 1e-20
) # features could be constant, e.g. voiceless speech
return feat
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
#!/usr/bin/env python
''' Example for sigproc.py '''
# pylint: skip-file
import scipy.io.wavfile as wav
from base import mfcc
from base import delta
from base import logfbank
if __name__ == '__main__':
(rate, sig) = wav.read("english.wav")
mfcc_feat = mfcc(sig, rate)
d_mfcc_feat = delta(mfcc_feat, 2)
fbank_feat = logfbank(sig, rate)
print(fbank_feat[1:3, :])