Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly
Appearance settings

Commit a350d7a

Browse filesBrowse files
jameslyonsjameslyons
authored
Merge pull request jameslyons#47 from tbfly/master
*): Adjust for coding style.
2 parents 3cd379e + 8dd2bfe commit a350d7a
Copy full SHA for a350d7a

2 files changed

+47-47Lines changed: 47 additions & 47 deletions

File tree

Expand file treeCollapse file tree
Open diff view settings
Filter options
Expand file treeCollapse file tree
Open diff view settings
Collapse file

‎python_speech_features/base.py‎

Copy file name to clipboardExpand all lines: python_speech_features/base.py
+25-25Lines changed: 25 additions & 25 deletions
Original file line numberDiff line numberDiff line change
@@ -19,12 +19,12 @@ def mfcc(signal,samplerate=16000,winlen=0.025,winstep=0.01,numcep=13,
1919
:param nfft: the FFT size. Default is 512.
2020
:param lowfreq: lowest band edge of mel filters. In Hz, default is 0.
2121
:param highfreq: highest band edge of mel filters. In Hz, default is samplerate/2
22-
:param preemph: apply preemphasis filter with preemph as coefficient. 0 is no filter. Default is 0.97.
23-
:param ceplifter: apply a lifter to final cepstral coefficients. 0 is no lifter. Default is 22.
22+
:param preemph: apply preemphasis filter with preemph as coefficient. 0 is no filter. Default is 0.97.
23+
:param ceplifter: apply a lifter to final cepstral coefficients. 0 is no lifter. Default is 22.
2424
:param appendEnergy: if this is true, the zeroth cepstral coefficient is replaced with the log of the total frame energy.
25-
: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
25+
: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
2626
:returns: A numpy array of size (NUMFRAMES by numcep) containing features. Each row holds 1 feature vector.
27-
"""
27+
"""
2828
feat,energy = fbank(signal,samplerate,winlen,winstep,nfilt,nfft,lowfreq,highfreq,preemph,winfunc)
2929
feat = numpy.log(feat)
3030
feat = dct(feat, type=2, axis=1, norm='ortho')[:,:numcep]
@@ -39,7 +39,7 @@ def fbank(signal,samplerate=16000,winlen=0.025,winstep=0.01,
3939
4040
:param signal: the audio signal from which to compute features. Should be an N*1 array
4141
:param samplerate: the samplerate of the signal we are working with.
42-
:param winlen: the length of the analysis window in seconds. Default is 0.025s (25 milliseconds)
42+
:param winlen: the length of the analysis window in seconds. Default is 0.025s (25 milliseconds)
4343
:param winstep: the step between successive windows in seconds. Default is 0.01s (10 milliseconds)
4444
:param nfilt: the number of filters in the filterbank, default 26.
4545
:param nfft: the FFT size. Default is 512.
@@ -49,18 +49,18 @@ def fbank(signal,samplerate=16000,winlen=0.025,winstep=0.01,
4949
: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
5050
:returns: 2 values. The first is a numpy array of size (NUMFRAMES by nfilt) containing features. Each row holds 1 feature vector. The
5151
second return value is the energy in each frame (total energy, unwindowed)
52-
"""
52+
"""
5353
highfreq= highfreq or samplerate/2
5454
signal = sigproc.preemphasis(signal,preemph)
5555
frames = sigproc.framesig(signal, winlen*samplerate, winstep*samplerate, winfunc)
5656
pspec = sigproc.powspec(frames,nfft)
5757
energy = numpy.sum(pspec,1) # this stores the total energy in each frame
5858
energy = numpy.where(energy == 0,numpy.finfo(float).eps,energy) # if energy is zero, we get problems with log
59-
59+
6060
fb = get_filterbanks(nfilt,nfft,samplerate,lowfreq,highfreq)
6161
feat = numpy.dot(pspec,fb.T) # compute the filterbank energies
6262
feat = numpy.where(feat == 0,numpy.finfo(float).eps,feat) # if feat is zero, we get problems with log
63-
63+
6464
return feat,energy
6565

6666
def logfbank(signal,samplerate=16000,winlen=0.025,winstep=0.01,
@@ -69,15 +69,15 @@ def logfbank(signal,samplerate=16000,winlen=0.025,winstep=0.01,
6969
7070
:param signal: the audio signal from which to compute features. Should be an N*1 array
7171
:param samplerate: the samplerate of the signal we are working with.
72-
:param winlen: the length of the analysis window in seconds. Default is 0.025s (25 milliseconds)
73-
:param winstep: the step between successive windows in seconds. Default is 0.01s (10 milliseconds)
72+
:param winlen: the length of the analysis window in seconds. Default is 0.025s (25 milliseconds)
73+
:param winstep: the step between successive windows in seconds. Default is 0.01s (10 milliseconds)
7474
:param nfilt: the number of filters in the filterbank, default 26.
7575
:param nfft: the FFT size. Default is 512.
7676
:param lowfreq: lowest band edge of mel filters. In Hz, default is 0.
7777
:param highfreq: highest band edge of mel filters. In Hz, default is samplerate/2
78-
:param preemph: apply preemphasis filter with preemph as coefficient. 0 is no filter. Default is 0.97.
79-
:returns: A numpy array of size (NUMFRAMES by nfilt) containing features. Each row holds 1 feature vector.
80-
"""
78+
:param preemph: apply preemphasis filter with preemph as coefficient. 0 is no filter. Default is 0.97.
79+
:returns: A numpy array of size (NUMFRAMES by nfilt) containing features. Each row holds 1 feature vector.
80+
"""
8181
feat,energy = fbank(signal,samplerate,winlen,winstep,nfilt,nfft,lowfreq,highfreq,preemph)
8282
return numpy.log(feat)
8383

@@ -88,36 +88,36 @@ def ssc(signal,samplerate=16000,winlen=0.025,winstep=0.01,
8888
8989
:param signal: the audio signal from which to compute features. Should be an N*1 array
9090
:param samplerate: the samplerate of the signal we are working with.
91-
:param winlen: the length of the analysis window in seconds. Default is 0.025s (25 milliseconds)
92-
:param winstep: the step between successive windows in seconds. Default is 0.01s (10 milliseconds)
91+
:param winlen: the length of the analysis window in seconds. Default is 0.025s (25 milliseconds)
92+
:param winstep: the step between successive windows in seconds. Default is 0.01s (10 milliseconds)
9393
:param nfilt: the number of filters in the filterbank, default 26.
9494
:param nfft: the FFT size. Default is 512.
9595
:param lowfreq: lowest band edge of mel filters. In Hz, default is 0.
9696
:param highfreq: highest band edge of mel filters. In Hz, default is samplerate/2
9797
:param preemph: apply preemphasis filter with preemph as coefficient. 0 is no filter. Default is 0.97.
9898
: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
99-
:returns: A numpy array of size (NUMFRAMES by nfilt) containing features. Each row holds 1 feature vector.
100-
"""
99+
:returns: A numpy array of size (NUMFRAMES by nfilt) containing features. Each row holds 1 feature vector.
100+
"""
101101
highfreq= highfreq or samplerate/2
102102
signal = sigproc.preemphasis(signal,preemph)
103103
frames = sigproc.framesig(signal, winlen*samplerate, winstep*samplerate, winfunc)
104104
pspec = sigproc.powspec(frames,nfft)
105105
pspec = numpy.where(pspec == 0,numpy.finfo(float).eps,pspec) # if things are all zeros we get problems
106-
106+
107107
fb = get_filterbanks(nfilt,nfft,samplerate,lowfreq,highfreq)
108108
feat = numpy.dot(pspec,fb.T) # compute the filterbank energies
109109
R = numpy.tile(numpy.linspace(1,samplerate/2,numpy.size(pspec,1)),(numpy.size(pspec,0),1))
110-
110+
111111
return numpy.dot(pspec*R,fb.T) / feat
112-
112+
113113
def hz2mel(hz):
114114
"""Convert a value in Hertz to Mels
115115
116116
:param hz: a value in Hz. This can also be a numpy array, conversion proceeds element-wise.
117117
:returns: a value in Mels. If an array was passed in, an identical sized array is returned.
118118
"""
119119
return 2595 * numpy.log10(1+hz/700.)
120-
120+
121121
def mel2hz(mel):
122122
"""Convert a value in Mels to Hertz
123123
@@ -139,7 +139,7 @@ def get_filterbanks(nfilt=20,nfft=512,samplerate=16000,lowfreq=0,highfreq=None):
139139
"""
140140
highfreq= highfreq or samplerate/2
141141
assert highfreq <= samplerate/2, "highfreq is greater than samplerate/2"
142-
142+
143143
# compute points evenly spaced in mels
144144
lowmel = hz2mel(lowfreq)
145145
highmel = hz2mel(highfreq)
@@ -154,12 +154,12 @@ def get_filterbanks(nfilt=20,nfft=512,samplerate=16000,lowfreq=0,highfreq=None):
154154
fbank[j,i] = (i - bin[j]) / (bin[j+1]-bin[j])
155155
for i in range(int(bin[j+1]), int(bin[j+2])):
156156
fbank[j,i] = (bin[j+2]-i) / (bin[j+2]-bin[j+1])
157-
return fbank
158-
157+
return fbank
158+
159159
def lifter(cepstra, L=22):
160160
"""Apply a cepstral lifter the the matrix of cepstra. This has the effect of increasing the
161161
magnitude of the high frequency DCT coeffs.
162-
162+
163163
:param cepstra: the matrix of mel-cepstra, will be numframes * numcep in size.
164164
:param L: the liftering coefficient to use. Default is 22. L <= 0 disables lifter.
165165
"""
Collapse file

‎python_speech_features/sigproc.py‎

Copy file name to clipboardExpand all lines: python_speech_features/sigproc.py
+22-22Lines changed: 22 additions & 22 deletions
Original file line numberDiff line numberDiff line change
@@ -53,68 +53,68 @@ def deframesig(frames,siglen,frame_len,frame_step,winfunc=lambda x:numpy.ones((x
5353
frame_step = round_half_up(frame_step)
5454
numframes = numpy.shape(frames)[0]
5555
assert numpy.shape(frames)[1] == frame_len, '"frames" matrix is wrong size, 2nd dim is not equal to frame_len'
56-
56+
5757
indices = numpy.tile(numpy.arange(0,frame_len),(numframes,1)) + numpy.tile(numpy.arange(0,numframes*frame_step,frame_step),(frame_len,1)).T
5858
indices = numpy.array(indices,dtype=numpy.int32)
59-
padlen = (numframes-1)*frame_step + frame_len
60-
59+
padlen = (numframes-1)*frame_step + frame_len
60+
6161
if siglen <= 0: siglen = padlen
62-
62+
6363
rec_signal = numpy.zeros((padlen,))
6464
window_correction = numpy.zeros((padlen,))
6565
win = winfunc(frame_len)
66-
66+
6767
for i in range(0,numframes):
6868
window_correction[indices[i,:]] = window_correction[indices[i,:]] + win + 1e-15 #add a little bit so it is never zero
6969
rec_signal[indices[i,:]] = rec_signal[indices[i,:]] + frames[i,:]
70-
70+
7171
rec_signal = rec_signal/window_correction
7272
return rec_signal[0:siglen]
73-
73+
7474
def magspec(frames,NFFT):
75-
"""Compute the magnitude spectrum of each frame in frames. If frames is an NxD matrix, output will be Nx(NFFT/2+1).
75+
"""Compute the magnitude spectrum of each frame in frames. If frames is an NxD matrix, output will be Nx(NFFT/2+1).
7676
7777
:param frames: the array of frames. Each row is a frame.
78-
:param NFFT: the FFT length to use. If NFFT > frame_len, the frames are zero-padded.
78+
:param NFFT: the FFT length to use. If NFFT > frame_len, the frames are zero-padded.
7979
:returns: If frames is an NxD matrix, output will be Nx(NFFT/2+1). Each row will be the magnitude spectrum of the corresponding frame.
80-
"""
81-
if numpy.shape(frames)[1] > NFFT:
80+
"""
81+
if numpy.shape(frames)[1] > NFFT:
8282
logging.warn('frame length (%d) is greater than FFT size (%d), frame will be truncated. Increase NFFT to avoid.', numpy.shape(frames)[1], NFFT)
8383
complex_spec = numpy.fft.rfft(frames,NFFT)
8484
return numpy.absolute(complex_spec)
85-
85+
8686
def powspec(frames,NFFT):
87-
"""Compute the power spectrum of each frame in frames. If frames is an NxD matrix, output will be Nx(NFFT/2+1).
87+
"""Compute the power spectrum of each frame in frames. If frames is an NxD matrix, output will be Nx(NFFT/2+1).
8888
8989
:param frames: the array of frames. Each row is a frame.
90-
:param NFFT: the FFT length to use. If NFFT > frame_len, the frames are zero-padded.
90+
:param NFFT: the FFT length to use. If NFFT > frame_len, the frames are zero-padded.
9191
:returns: If frames is an NxD matrix, output will be Nx(NFFT/2+1). Each row will be the power spectrum of the corresponding frame.
92-
"""
92+
"""
9393
return 1.0/NFFT * numpy.square(magspec(frames,NFFT))
94-
94+
9595
def logpowspec(frames,NFFT,norm=1):
96-
"""Compute the log power spectrum of each frame in frames. If frames is an NxD matrix, output will be Nx(NFFT/2+1).
96+
"""Compute the log power spectrum of each frame in frames. If frames is an NxD matrix, output will be Nx(NFFT/2+1).
9797
9898
:param frames: the array of frames. Each row is a frame.
99-
:param NFFT: the FFT length to use. If NFFT > frame_len, the frames are zero-padded.
99+
:param NFFT: the FFT length to use. If NFFT > frame_len, the frames are zero-padded.
100100
:param norm: If norm=1, the log power spectrum is normalised so that the max value (across all frames) is 0.
101101
:returns: If frames is an NxD matrix, output will be Nx(NFFT/2+1). Each row will be the log power spectrum of the corresponding frame.
102-
"""
102+
"""
103103
ps = powspec(frames,NFFT);
104104
ps[ps<=1e-30] = 1e-30
105105
lps = 10*numpy.log10(ps)
106106
if norm:
107107
return lps - numpy.max(lps)
108108
else:
109109
return lps
110-
110+
111111
def preemphasis(signal,coeff=0.95):
112112
"""perform preemphasis on the input signal.
113-
113+
114114
:param signal: The signal to filter.
115115
:param coeff: The preemphasis coefficient. 0 is no filter, default is 0.95.
116116
:returns: the filtered signal.
117-
"""
117+
"""
118118
return numpy.append(signal[0],signal[1:]-coeff*signal[:-1])
119119

120120

0 commit comments

Comments
0 (0)
Morty Proxy This is a proxified and sanitized view of the page, visit original site.