def ssc(signal,samplerate=16000,winlen=0.025,winstep=0.01,
nfilt=26,nfft=512,lowfreq=0,highfreq=None,preemph=0.97,
winfunc=lambda x:np.ones((x,))):
"""Compute Spectral Subband Centroid features from an audio signal.
:param signal: the audio signal from which to compute features. Should be an N*1 array
:param samplerate: the sample rate of the signal we are working with, in Hz.
:param winlen: the length of the analysis window in seconds. Default is 0.025s (25 milliseconds)
:param winstep: the step between successive windows in seconds. Default is 0.01s (10 milliseconds)
:param nfilt: the number of filters in the filterbank, default 26.
:param nfft: the FFT size. Default is 512.
:param lowfreq: lowest band edge of mel filters. In Hz, default is 0.
:param highfreq: highest band edge of mel filters. In Hz, default is samplerate/2
:param preemph: apply preemphasis filter with preemph as coefficient. 0 is no filter. Default is 0.97.
:param winfunc: the analysis window to apply to each frame. By default no window is applied. You can use numpy window functions here e.g. winfunc=numpy.hamming
:returns: A numpy array of size (NUMFRAMES by nfilt) containing features. Each row holds 1 feature vector.
"""
highfreq= highfreq or samplerate/2
signal = sigproc.preemphasis(signal,preemph)
frames = sigproc.framesig(signal, winlen*samplerate, winstep*samplerate, winfunc)
pspec = sigproc.powspec(frames,nfft)
pspec = np.where(pspec == 0,np.finfo(float).eps,pspec) # if things are all zeros we get problems
fb = get_filterbanks(nfilt,nfft,samplerate,lowfreq,highfreq)
feat = np.dot(pspec,fb.T) # compute the filterbank energies
R = np.tile(np.linspace(1,samplerate/2,np.size(pspec,1)),(np.size(pspec,0),1))
return np.dot(pspec*R,fb.T) / feat
def ssc(signal,samplerate=16000,winlen=0.025,winstep=0.01,
nfilt=26,nfft=512,lowfreq=0,highfreq=None,preemph=0.97):
"""Compute Spectral Subband Centroid features from an audio signal.
:param signal: the audio signal from which to compute features. Should be an N*1 array
:param samplerate: the samplerate of the signal we are working with.
:param winlen: the length of the analysis window in seconds. Default is 0.025s (25 milliseconds)
:param winstep: the step between seccessive windows in seconds. Default is 0.01s (10 milliseconds)
:param nfilt: the number of filters in the filterbank, default 20.
:param nfft: the FFT size. Default is 512.
:param lowfreq: lowest band edge of mel filters. In Hz, default is 0.
:param highfreq: highest band edge of mel filters. In Hz, default is samplerate/2
:param preemph: apply preemphasis filter with preemph as coefficient. 0 is no filter. Default is 0.97.
:returns: A numpy array of size (NUMFRAMES by nfilt) containing features. Each row holds 1 feature vector.
"""
highfreq= highfreq or samplerate/2
signal = sigproc.preemphasis(signal,preemph)
frames = sigproc.framesig(signal, winlen*samplerate, winstep*samplerate)
pspec = sigproc.powspec(frames,nfft)
fb = get_filterbanks(nfilt,nfft,samplerate)
feat = numpy.dot(pspec,fb.T) # compute the filterbank energies
R = numpy.tile(numpy.linspace(1,samplerate/2,numpy.size(pspec,1)),(numpy.size(pspec,0),1))
return numpy.dot(pspec*R,fb.T) / feat
def ssc(signal,
samplerate=16000,
winlen=0.025,
winstep=0.01,
nfilt=26,
nfft=512,
lowfreq=0,
highfreq=None,
preemph=0.97,
winfunc=lambda x: np.ones((x, ))):
highfreq = highfreq or samplerate / 2
signal = sigproc.preemphasis(signal, preemph)
frames = sigproc.framesig(signal, winlen * samplerate,
winstep * samplerate, winfunc)
pspec = sigproc.powspec(frames, nfft)
pspec = np.where(pspec == 0,
np.finfo(float).eps,
pspec) # if things are all zeros we get problems
fb = get_filterbanks(nfilt, nfft, samplerate, lowfreq, highfreq)
feat = np.dot(pspec, fb.T) # compute the filterbank energies
R = np.tile(np.linspace(1, samplerate / 2, np.size(pspec, 1)),
(np.size(pspec, 0), 1))
return np.dot(pspec * R, fb.T) / feat
def fbank(signal,samplerate=16000,winlen=0.025,winstep=0.01,
nfilt=26,nfft=512,lowfreq=0,highfreq=None,preemph=0.97):
"""Compute Mel-filterbank energy features from an audio signal.
:param signal: the audio signal from which to compute features. Should be an N*1 array
:param samplerate: the samplerate of the signal we are working with.
:param winlen: the length of the analysis window in seconds. Default is 0.025s (25 milliseconds)
:param winstep: the step between seccessive windows in seconds. Default is 0.01s (10 milliseconds)
:param nfilt: the number of filters in the filterbank, default 20.
:param nfft: the FFT size. Default is 512.
:param lowfreq: lowest band edge of mel filters. In Hz, default is 0.
:param highfreq: highest band edge of mel filters. In Hz, default is samplerate/2
:param preemph: apply preemphasis filter with preemph as coefficient. 0 is no filter. Default is 0.97.
:returns: 2 values. The first is a numpy array of size (NUMFRAMES by nfilt) containing features. Each row holds 1 feature vector. The
second return value is the energy in each frame (total energy, unwindowed)
"""
highfreq= highfreq or samplerate/2
signal = sigproc.preemphasis(signal,preemph)
frames = sigproc.framesig(signal, winlen*samplerate, winstep*samplerate)
pspec = sigproc.powspec(frames,nfft)
energy = numpy.sum(pspec,1) # this stores the total energy in each frame
fb = get_filterbanks(nfilt,nfft,samplerate)
feat = numpy.dot(pspec,fb.T) # compute the filterbank energies
return feat,energy
def fbank(signal,
samplerate=16000,
winlen=0.025,
winstep=0.01,
nfilt=26,
nfft=512,
lowfreq=0,
highfreq=None,
preemph=0.97,
winfunc=lambda x: np.ones((x, ))):
highfreq = highfreq or samplerate / 2
signal = sigproc.preemphasis(signal, preemph)
frames = sigproc.framesig(signal, winlen * samplerate,
winstep * samplerate, winfunc)
pspec = sigproc.powspec(frames, nfft)
energy = np.sum(pspec, 1) # this stores the total energy in each frame
energy = np.where(
energy == 0,
np.finfo(float).eps,
energy) # if energy is zero, we get problems with log
fb = get_filterbanks(nfilt, nfft, samplerate, lowfreq, highfreq)
feat = np.dot(pspec, fb.T) # compute the filterbank energies
feat = np.where(feat == 0,
np.finfo(float).eps,
feat) # if feat is zero, we get problems with log
return feat, energy
def fbank(signal,samplerate=16000,winlen=0.025,winstep=0.01,
nfilt=26,nfft=512,lowfreq=0,highfreq=None,preemph=0.97):
"""Compute Mel-filterbank energy features from an audio signal.
:param signal: the audio signal from which to compute features. Should be an N*1 array
:param samplerate: the samplerate of the signal we are working with.
:param winlen: the length of the analysis window in seconds. Default is 0.025s (25 milliseconds)
:param winstep: the step between successive windows in seconds. Default is 0.01s (10 milliseconds)
:param nfilt: the number of filters in the filterbank, default 26.
:param nfft: the FFT size. Default is 512.
:param lowfreq: lowest band edge of mel filters. In Hz, default is 0.
:param highfreq: highest band edge of mel filters. In Hz, default is samplerate/2
:param preemph: apply preemphasis filter with preemph as coefficient. 0 is no filter. Default is 0.97.
:returns: 2 values. The first is a numpy array of size (NUMFRAMES by nfilt) containing features. Each row holds 1 feature vector. The
second return value is the energy in each frame (total energy, unwindowed)
"""
highfreq= highfreq or samplerate/2
signal = sigproc.preemphasis(signal,preemph)
frames = sigproc.framesig(signal, winlen*samplerate, winstep*samplerate)
pspec = sigproc.powspec(frames,nfft)
energy = numpy.sum(pspec,1) # this stores the total energy in each frame
energy = numpy.where(energy == 0,numpy.finfo(float).eps,energy) # if energy is zero, we get problems with log
fb = get_filterbanks(nfilt,nfft,samplerate,lowfreq,highfreq)
feat = numpy.dot(pspec,fb.T) # compute the filterbank energies
feat = numpy.where(feat == 0,numpy.finfo(float).eps,feat) # if feat is zero, we get problems with log
return feat,energy
def ssc(signal,samplerate=16000,winlen=0.025,winstep=0.01,
nfilt=26,nfft=512,lowfreq=0,highfreq=None,preemph=0.97):
"""Compute Spectral Subband Centroid features from an audio signal.
:param signal: the audio signal from which to compute features. Should be an N*1 array
:param samplerate: the samplerate of the signal we are working with.
:param winlen: the length of the analysis window in seconds. Default is 0.025s (25 milliseconds)
:param winstep: the step between successive windows in seconds. Default is 0.01s (10 milliseconds)
:param nfilt: the number of filters in the filterbank, default 26.
:param nfft: the FFT size. Default is 512.
:param lowfreq: lowest band edge of mel filters. In Hz, default is 0.
:param highfreq: highest band edge of mel filters. In Hz, default is samplerate/2
:param preemph: apply preemphasis filter with preemph as coefficient. 0 is no filter. Default is 0.97.
:returns: A numpy array of size (NUMFRAMES by nfilt) containing features. Each row holds 1 feature vector.
"""
highfreq= highfreq or samplerate/2
signal = sigproc.preemphasis(signal,preemph)
frames = sigproc.framesig(signal, winlen*samplerate, winstep*samplerate)
pspec = sigproc.powspec(frames,nfft)
pspec = numpy.where(pspec == 0,numpy.finfo(float).eps,pspec) # if things are all zeros we get problems
fb = get_filterbanks(nfilt,nfft,samplerate,lowfreq,highfreq)
feat = numpy.dot(pspec,fb.T) # compute the filterbank energies
R = numpy.tile(numpy.linspace(1,samplerate/2,numpy.size(pspec,1)),(numpy.size(pspec,0),1))
return numpy.dot(pspec*R,fb.T) / feat
def powspec(signal, samplerate, conf):
"""
Compute squared magnitude spectrogram features from an audio signal.
Args:
signal: the audio signal from which to compute features. Should be an
N*1 array
samplerate: the samplerate of the signal we are working with.
conf: feature configuration
Returns:
A numpy array of size (NUMFRAMES by numfreq) containing features. Each
row holds 1 feature vector, a numpy vector containing the magnitude
spectrum of the corresponding frame
"""
signal = sigproc.preemphasis(signal, float(conf['preemph']))
winfunc = _get_winfunc(conf['winfunc'])
frames = sigproc.framesig(signal, float(conf['winlen'])*samplerate,
float(conf['winstep'])*samplerate,
winfunc)
powspec = sigproc.powspec(frames, int(conf['nfft']))
return powspec
def fbank(signal,samplerate=16000,winlen=0.025,winstep=0.01,
nfilt=26,nfft=512,lowfreq=0,highfreq=None,preemph=0.97,
winfunc=lambda x:np.ones((x,))):
"""Compute Mel-filterbank energy features from an audio signal.
:param signal: the audio signal from which to compute features. Should be an N*1 array
:param samplerate: the sample rate of the signal we are working with, in Hz.
:param winlen: the length of the analysis window in seconds. Default is 0.025s (25 milliseconds)
:param winstep: the step between successive windows in seconds. Default is 0.01s (10 milliseconds)
:param nfilt: the number of filters in the filterbank, default 26.
:param nfft: the FFT size. Default is 512.
:param lowfreq: lowest band edge of mel filters. In Hz, default is 0.
:param highfreq: highest band edge of mel filters. In Hz, default is samplerate/2
:param preemph: apply preemphasis filter with preemph as coefficient. 0 is no filter. Default is 0.97.
:param winfunc: the analysis window to apply to each frame. By default no window is applied. You can use numpy window functions here e.g. winfunc=numpy.hamming
:returns: 2 values. The first is a numpy array of size (NUMFRAMES by nfilt) containing features. Each row holds 1 feature vector. The
second return value is the energy in each frame (total energy, unwindowed)
"""
highfreq= highfreq or samplerate/2
signal = sigproc.preemphasis(signal,preemph)
frames = sigproc.framesig(signal, winlen*samplerate, winstep*samplerate, winfunc)
pspec = sigproc.powspec(frames,nfft)
energy = np.sum(pspec,1) # this stores the total energy in each frame
energy = np.where(energy == 0,np.finfo(float).eps,energy) # if energy is zero, we get problems with log
fb = get_filterbanks(nfilt,nfft,samplerate,lowfreq,highfreq)
feat = np.dot(pspec,fb.T) # compute the filterbank energies
feat = np.where(feat == 0,np.finfo(float).eps,feat) # if feat is zero, we get problems with log
return feat,energy
def do_mfcc(self, fft_n, mffc_ch):
""" produces magnitude-> pow.spec on a freq scale """
self.nfft = fft_n
self.nmffc = mffc_ch
self.mag = sigproc.magspec(self.samples, self.nfft)
self.powspec2 = np.square(self.mag)
self.powspec1 = sigproc.powspec(self.samples,self.nfft)
self.bank = get_filterbanks(nfilt=self.nmffc, nfft=self.nfft, samplerate=self.hz)
self.mfc = np.dot(self.powspec1, self.bank.T)
return self.mfc
def fbank(signal,samplerate=16000,winlen=0.025,winstep=0.01,
nfilt=26,nfft=512,lowfreq=0,highfreq=None,preemph=0.97):
"""Compute Mel-filterbank energy features from an audio signal.
:param signal: the audio signal from which to compute features. Should be an N*1 array
:param samplerate: the samplerate of the signal we are working with.
:param winlen: the length of the analysis window in seconds. Default is 0.025s (25 milliseconds)
:param winstep: the step between successive windows in seconds. Default is 0.01s (10 milliseconds)
:param nfilt: the number of filters in the filterbank, default 26.
:param nfft: the FFT size. Default is 512.
:param lowfreq: lowest band edge of mel filters. In Hz, default is 0.
:param highfreq: highest band edge of mel filters. In Hz, default is samplerate/2
:param preemph: apply preemphasis filter with preemph as coefficient. 0 is no filter. Default is 0.97.
:returns: 2 values. The first is a numpy array of size (NUMFRAMES by nfilt) containing features. Each row holds 1 feature vector. The
second return value is the energy in each frame (total energy, unwindowed)
"""
highfreq= highfreq or samplerate/2
print "preemph %s"%(preemph)
signal = sigproc.preemphasis(signal,preemph)
frames = sigproc.framesig(signal, winlen*samplerate, winstep*samplerate)
matchframes(frames[0], frames[1])
pspec = sigproc.powspec(frames,nfft)
energy = pylab.sum(pspec,1) # this stores the total energy in each frame
energy = pylab.where(energy == 0, pylab.finfo(float).eps, energy) # if energy is zero, we get problems with log
fb = get_filterbanks(nfilt, nfft, samplerate, lowfreq, highfreq)
print "len(fb) %s"%(len(fb))
colour = "k-"
for i in range(len(fb)):
if colour == "k-":
colour = "r-"
else:
colour = "k-"
startedplot = False
midpoint = 0
for j in range(len(fb[i])):
if fb[i][j] > 0:
if startedplot == False:
startedplot = j
if j > 0:
pylab.plot([j-1, j], [fb[i][j-1], fb[i][j]], colour)
if fb[i][j] == 1.0:
midpoint = j
else:
if not startedplot == False:
pylab.plot([j-1, j], [fb[i][j-1], 0], colour)
try:
print "slope to midpoint %.3f, slope from midpoint %.3f"%(1.0/float(midpoint-startedplot), 1.0/float(midpoint-j+1))
except:
pass
break
pylab.show()
feat = pylab.dot(pspec, fb.T) # compute the filterbank energies
feat = pylab.where(feat == 0, pylab.finfo(float).eps, feat) # if feat is zero, we get problems with log
return feat, energy
def ssc(signal,samplerate=16000,winlen=0.025,winstep=0.01,
nfilt=26,nfft=512,lowfreq=0,highfreq=None,preemph=0.97):
highfreq= highfreq or samplerate/2
signal = sigproc.preemphasis(signal,preemph)
frames = sigproc.framesig(signal, winlen*samplerate, winstep*samplerate)
pspec = sigproc.powspec(frames,nfft)
pspec = numpy.where(pspec == 0,numpy.finfo(float).eps,pspec) # if things are all zeros we get problems
fb = get_filterbanks(nfilt,nfft,samplerate,lowfreq,highfreq)
feat = numpy.dot(pspec,fb.T) # compute the filterbank energies
R = numpy.tile(numpy.linspace(1,samplerate/2,numpy.size(pspec,1)),(numpy.size(pspec,0),1))
return numpy.dot(pspec*R,fb.T) / feat
def fbank(signal,samplerate=16000,winlen=0.025,winstep=0.01,
nfilt=26,nfft=512,lowfreq=0,highfreq=None,preemph=0.97):
highfreq= highfreq or samplerate/2
signal = sigproc.preemphasis(signal,preemph)
frames = sigproc.framesig(signal, winlen*samplerate, winstep*samplerate)
pspec = sigproc.powspec(frames,nfft)
energy = numpy.sum(pspec,1) # this stores the total energy in each frame
energy = numpy.where(energy == 0,numpy.finfo(float).eps,energy) # if energy is zero, we get problems with log
fb = get_filterbanks(nfilt,nfft,samplerate,lowfreq,highfreq)
feat = numpy.dot(pspec,fb.T) # compute the filterbank energies
feat = numpy.where(feat == 0,numpy.finfo(float).eps,feat) # if feat is zero, we get problems with log
return feat,energy
def fbank(signal,
samplerate=16000,
winlen=0.025,
winstep=0.01,
nfilt=26,
nfft=512,
lowfreq=0,
highfreq=None,
preemph=0.95):
"""
Compute Mel-filterbank energy features from an audio signal.
:param signal: the audio signal from which to compute features. Should be an N*1 array
:param samplerate: the samplerate of the signal we are working with.
:param winlen: the length of the analysis window in seconds. Default is 0.025s (25 milliseconds)
:param winstep: the step between successive windows in seconds. Default is 0.01s (10 milliseconds)
:param nfilt: the number of filters in the filterbank, default 26.
:param nfft: the FFT size. Default is 512.
:param lowfreq: lowest band edge of mel filters. In Hz, default is 0.
:param highfreq: highest band edge of mel filters. In Hz, default is samplerate/2
:param preemph: apply preemphasis filter with preemph as coefficient. 0 is no filter. Default is 0.95.
:returns: 2 values. The first is a numpy array of size (NUMFRAMES by nfilt) containing features. Each row holds 1 feature vector. The
second return value is the energy in each frame (total energy, unwindowed)
"""
highfreq = highfreq or samplerate / 2
signal = sigproc.preemphasis(signal, preemph)
# print type(signal[0])
frames = sigproc.framesig(signal,
winlen * samplerate,
winstep * samplerate,
winfunc=hamming_window)
powspec = sigproc.powspec(frames, nfft)
# numpy.savetxt("result.txt", powspec, delimiter=",")
energy = numpy.sum(powspec,
1) # this stores the total energy in each frame
energy = numpy.where(
energy == 0,
numpy.finfo(float).eps, energy
) # if energy is zero, we get problems with log, use numpy.finfo(float).eps to replace 0
filterbanks = get_filterbanks(nfilt, nfft, samplerate, lowfreq, highfreq)
# print powspec.shape, filterbanks.shape
feat = numpy.dot(powspec, filterbanks.T) # compute the filterbank energies
feat = numpy.where(feat == 0,
numpy.finfo(float).eps,
feat) # if feat is zero, we get problems with logs
# print feat.shape
return feat, energy
def fbank(signal, samplerate, conf):
'''
Compute fbank features from an audio signal.
从一个声音信号中计算fbank特征向量
Args:
参数：
signal: the audio signal from which to compute features. Should be an
N*1 array
要计算特征的声音信号，一个N*1维的数组
samplerate: the samplerate of the signal we are working with.
要处理信号的采样率
conf: feature configuration
特征的配置
Returns:
返回值：
A numpy array of size (NUMFRAMES by nfilt) containing features, a numpy
vector containing the signal energy
返回一个包含特征向量的numpy数组，一个包含信号能量的numpy向量
'''
highfreq = int(conf['highfreq'])
if highfreq < 0:
highfreq = samplerate/2
signal = sigproc.preemphasis(signal, float(conf['preemph']))
frames = sigproc.framesig(signal, float(conf['winlen'])*samplerate,
float(conf['winstep'])*samplerate)
pspec = sigproc.powspec(frames, int(conf['nfft']))
# this stores the total energy in each frame
energy = numpy.sum(pspec, 1)
# if energy is zero, we get problems with log
energy = numpy.where(energy == 0, numpy.finfo(float).eps, energy)
filterbank = get_filterbanks(int(conf['nfilt']), int(conf['nfft']),
samplerate, int(conf['lowfreq']), highfreq)
# compute the filterbank energies
feat = numpy.dot(pspec, filterbank.T)
# if feat is zero, we get problems with log
feat = numpy.where(feat == 0, numpy.finfo(float).eps, feat)
return feat, energy
def fbankVTLP(signal,
samplerate=16000,
winlen=0.025,
winstep=0.01,
nfilt=26,
nfft=512,
lowfreq=0,
highfreq=None,
preemph=0.97,
appendEnergy=False,
alpha=1.0):
"""Compute Mel-filterbank energy features from an audio signal.
:param signal: the audio signal from which to compute features. Should be an N*1 array
:param samplerate: the samplerate of the signal we are working with.
:param winlen: the length of the analysis window in seconds. Default is 0.025s (25 milliseconds)
:param winstep: the step between successive windows in seconds. Default is 0.01s (10 milliseconds)
:param nfilt: the number of filters in the filterbank, default 26.
:param nfft: the FFT size. Default is 512.
:param lowfreq: lowest band edge of mel filters. In Hz, default is 0.
:param highfreq: highest band edge of mel filters. In Hz, default is samplerate/2
:param preemph: apply preemphasis filter with preemph as coefficient. 0 is no filter. Default is 0.97.
:returns: 2 values. The first is a numpy array of size (NUMFRAMES by nfilt) containing features. Each row holds 1 feature vector. The
second return value is the energy in each frame (total energy, unwindowed)
"""
highfreq = highfreq or samplerate / 2
signal = sigproc.preemphasis(signal, preemph)
frames = sigproc.framesig(signal, winlen * samplerate,
winstep * samplerate)
pspec = sigproc.powspec(frames, nfft)
energy = numpy.sum(pspec, 1) # this stores the total energy in each frame
energy = numpy.where(energy == 0,
numpy.finfo(float).eps,
energy) # if energy is zero, we get problems with log
fb = get_filterbanksVTLP(nfilt, nfft, samplerate, lowfreq, highfreq, alpha)
feat = numpy.dot(pspec, fb.T) # compute the filterbank energies
feat = numpy.where(feat == 0,
numpy.finfo(float).eps,
feat) # if feat is zero, we get problems with log
if appendEnergy:
feat = numpy.c_[feat, numpy.log(
energy
)] # replace first cepstral coefficient with log of frame energy
return feat, energy
def fbank(signal, samplerate, conf):
"""
Compute fbank features from an audio signal.
Args:
signal: the audio signal from which to compute features. Should be an
N*1 array
samplerate: the samplerate of the signal we are working with.
conf: feature configuration
Returns:
A numpy array of size (NUMFRAMES by nfilt) containing features, a numpy
vector containing the signal energy
"""
raise BaseException('Not yet implemented')
highfreq = int(conf['highfreq'])
if highfreq < 0:
highfreq = samplerate / 2
signal = sigproc.preemphasis(signal, float(conf['preemph']))
frames = sigproc.framesig(signal,
float(conf['winlen']) * samplerate,
float(conf['winstep']) * samplerate)
pspec = sigproc.powspec(frames, int(conf['nfft']))
# this stores the total energy in each frame
energy = numpy.sum(pspec, 1)
# if energy is zero, we get problems with log
energy = numpy.where(energy == 0, numpy.finfo(float).eps, energy)
filterbank = get_filterbanks(int(conf['nfilt']), int(conf['nfft']),
samplerate, int(conf['lowfreq']), highfreq)
# compute the filterbank energies
feat = numpy.dot(pspec, filterbank.T)
# if feat is zero, we get problems with log
feat = numpy.where(feat == 0, numpy.finfo(float).eps, feat)
return feat, energy
def ssc(signal, samplerate, conf):
'''
Compute ssc features from an audio signal.
Args:
signal: the audio signal from which to compute features. Should be an
N*1 array
samplerate: the samplerate of the signal we are working with.
conf: feature configuration
Returns:
A numpy array of size (NUMFRAMES by nfilt) containing features, a numpy
vector containing the signal log-energy
'''
highfreq = int(conf['highfreq'])
if highfreq < 0:
highfreq = samplerate / 2
signal = sigproc.preemphasis(signal, float(conf['preemph']))
frames = sigproc.framesig(signal,
float(conf['winlen']) * samplerate,
float(conf['winstep']) * samplerate)
pspec = sigproc.powspec(frames, int(conf['nfft']))
# this stores the total energy in each frame
energy = numpy.sum(pspec, 1)
# if energy is zero, we get problems with log
energy = numpy.where(energy == 0, numpy.finfo(float).eps, energy)
filterbank = get_filterbanks(int(conf['nfilt']), int(conf['nfft']),
samplerate, int(conf['lowfreq']), highfreq)
# compute the filterbank energies
feat = numpy.dot(pspec, filterbank.T)
tiles = numpy.tile(numpy.linspace(1, samplerate / 2, numpy.size(pspec, 1)),
(numpy.size(pspec, 0), 1))
return numpy.dot(pspec * tiles, filterbank.T) / feat, numpy.log(energy)
