1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
|
#! /usr/bin/env python
# Last Change: Tue Nov 28 03:00 PM 2006 J
# TODO: make lpc work for any rank
from warnings import warn
import numpy as _N
import scipy.signal as _sig
from scipy.linalg import toeplitz, inv
from autocorr import autocorr_oneside_nofft
# use ctype to have fast lpc computation
import ctypes
from ctypes import c_uint, c_int
from numpy.ctypeslib import ndpointer, load_library
ctypes_major = int(ctypes.__version__.split('.')[0])
if ctypes_major < 1:
msg = "version of ctypes is %s, expected at least %s" \
% (ctypes.__version__, '1.0.1')
raise importerror(msg)
# load lpc
_lpc = load_library('gabsig.so', __file__)
#===============================
# define the functions with args
#===============================
arg1 = ndpointer(dtype = _N.float64, flags = 'C')
arg2 = c_uint
arg3 = c_uint
arg4 = ndpointer(dtype = _N.float64, flags = 'C')
arg5 = ndpointer(dtype = _N.float64, flags = 'C')
arg6 = ndpointer(dtype = _N.float64, flags = 'C')
_lpc.dbl_lpc.argtypes = [arg1, arg2, arg3, arg4, arg5, arg6]
_lpc.dbl_lpc.restype = c_int
arg1 = ndpointer(dtype = _N.float32, flags = 'C')
arg2 = c_uint
arg3 = c_uint
arg4 = ndpointer(dtype = _N.float32, flags = 'C')
arg5 = ndpointer(dtype = _N.float32, flags = 'C')
arg6 = ndpointer(dtype = _N.float32, flags = 'C')
_lpc.flt_lpc.argtypes = [arg1, arg2, arg3, arg4, arg5, arg6]
_lpc.flt_lpc.restype = c_int
arg1 = ndpointer(dtype = _N.float64, flags = 'C')
arg2 = c_uint
arg3 = c_uint
arg4 = ndpointer(dtype = _N.float64, flags = 'C')
arg5 = ndpointer(dtype = _N.float64, flags = 'C')
arg6 = ndpointer(dtype = _N.float64, flags = 'C')
_lpc.dbl_levinson2d.argtypes = [arg1, arg2, arg3, arg4, arg5, arg6]
_lpc.dbl_levinson2d.restype = c_int
arg1 = ndpointer(dtype = _N.float32, flags = 'C')
arg2 = c_uint
arg3 = c_uint
arg4 = ndpointer(dtype = _N.float32, flags = 'C')
arg5 = ndpointer(dtype = _N.float32, flags = 'C')
arg6 = ndpointer(dtype = _N.float32, flags = 'C')
_lpc.flt_levinson2d.argtypes = [arg1, arg2, arg3, arg4, arg5, arg6]
_lpc.flt_levinson2d.restype = c_int
def lpc_ref(signal, order):
""" Return the order + 1 LPC coefficients
for the signal. This is just for reference, as it is using
the direct inversion of the toeplitz matrix, which is really slow"""
if signal.ndim > 1:
print "Warning, not tested for rank > 1"
p = order + 1
r = autocorr_oneside_nofft(signal, order) / signal.shape[0]
return _N.concatenate(([1.], _N.dot(inv(toeplitz(r[:-1])), -r[1:])))
def lpc(signal, order):
""" Return the order + 1 LPC coefficients
for the signal using levinson durbin algorithm """
if signal.ndim > 1:
warn("Warning, not tested for rank > 1")
if signal.size <= order:
raise RuntimeError("size is smaller than order !")
if signal.dtype == _N.float64:
coeff = _N.zeros((order+1), _N.float64)
kcoeff = _N.zeros((order), _N.float64)
err = _N.zeros((1), _N.float64)
st = _lpc.dbl_lpc(signal, signal.size, order, coeff, kcoeff, err)
elif signal.dtype == _N.float32:
coeff = _N.zeros((order+1), _N.float32)
kcoeff = _N.zeros((order), _N.float32)
err = _N.zeros((1), _N.float32)
st = _lpc.flt_lpc(signal, signal.size, order, coeff, kcoeff, err)
else:
raise TypeError("Sorry, only float32 and float64 supported")
if not (st == 0):
raise RuntimeError("Error while using levinson algo, returns err is %d", st)
return coeff, err, kcoeff
def lpcres_ref(signal, order, axis = 0):
return _sig.lfilter(lpc_ref(signal, order), 1., signal, axis = 0)
def lpcres(signal, order, axis = 0):
return _sig.lfilter(lpc(signal, order)[0], 1., signal, axis = 0)
def _lpc2_py(signal, order, axis = -1):
"""python implementation of lpc for rank 2., Do not use, for testing purpose only"""
if signal.ndim > 2:
raise NotImplemented("only for rank <=2")
if signal.ndim < 2:
return lpc(_N.require(signal, requirements = 'C'), order)
# For each array of direction axis, compute levinson durbin
if axis % 2 == 0:
# Prepare output arrays
coeff = _N.zeros((order+1, signal.shape[1]), signal.dtype)
kcoeff = _N.zeros((order, signal.shape[1]), signal.dtype)
err = _N.zeros(signal.shape[1], signal.dtype)
for i in range(signal.shape[1]):
coeff[:, i], err[i], kcoeff[:, i] = \
lpc(_N.require(signal[:, i], requirements = 'C'), order)
elif axis % 2 == 1:
# Prepare output arrays
coeff = _N.zeros((signal.shape[0], order+1), signal.dtype)
kcoeff = _N.zeros((signal.shape[0], order), signal.dtype)
err = _N.zeros(signal.shape[0], signal.dtype)
for i in range(signal.shape[0]):
coeff[i], err[i], kcoeff[i] = \
lpc(_N.require(signal[i], requirements = 'C'), order)
else:
raise RuntimeError("this should not happen, please fill a bug")
return coeff, err, kcoeff
def lpc2(signal, order, axis = -1):
""" Returns ar coeff, err and k coeff"""
sz = signal.shape[axis]
if order >= sz:
raise RuntimeError("order should be strictly smaller than the length of signal")
# rank 1
if signal.ndim < 2:
if signal.dtype == _N.float64:
coeff = _N.zeros((order+1), _N.float64)
kcoeff = _N.zeros((order), _N.float64)
err = _N.zeros((1), _N.float64)
st = _lpc.dbl_lpc(signal, signal.size, order, coeff, kcoeff, err)
elif signal.dtype == _N.float32:
coeff = _N.zeros((order+1), _N.float32)
kcoeff = _N.zeros((order), _N.float32)
err = _N.zeros((1), _N.float32)
st = _lpc.flt_lpc(signal, signal.size, order, coeff, kcoeff, err)
else:
raise TypeError("Sorry, only float32 and float64 supported")
if not (st == 0):
raise RuntimeError("Error while using levinson algo, returns err is %d", st)
return coeff, err, kcoeff
# rank 2
elif signal.ndim == 2:
# Compute biased autocorrelation up to lag = order
bias = 1. / sz
corr = bias * autocorr_oneside_nofft(signal, order, axis)
if axis % 2 == 0:
# we transpose to have a major row order
icorr = corr.T
icorr = _N.require(icorr, requirements = 'C')
at = _N.zeros((icorr.shape[0], order+1), icorr.dtype)
kt = _N.zeros((icorr.shape[0], order), icorr.dtype)
et = _N.zeros(icorr.shape[0], icorr.dtype)
if icorr.dtype == _N.float64:
_lpc.dbl_levinson2d(icorr, icorr.shape[0], icorr.shape[1], at, et, kt)
elif icorr.dtype == _N.float32:
_lpc.flt_levinson2d(icorr, icorr.shape[0], icorr.shape[1], at, et, kt)
else:
raise TypeError("Only float32 and float64 supported")
return at.T, et.T, kt.T
elif axis % 2 == 1:
icorr = _N.require(corr, requirements = 'C')
at = _N.zeros((icorr.shape[0], order+1), icorr.dtype)
kt = _N.zeros((icorr.shape[0], order), icorr.dtype)
et = _N.zeros(icorr.shape[0], icorr.dtype)
if icorr.dtype == _N.float64:
_lpc.dbl_levinson2d(icorr, icorr.shape[0], icorr.shape[1], at, et, kt)
elif icorr.dtype == _N.float32:
_lpc.flt_levinson2d(icorr, icorr.shape[0], icorr.shape[1], at, et, kt)
else:
raise TypeError("Only float32 and float64 supported")
return at, et, kt
else:
raise RuntimeError("This should not happen, this is a bug")
else:
raise RuntimeError("Sorry, only rank <= 2 supported for now")
def bench():
size = 256
nframes = 4000
order = 24
X = _N.random.randn(nframes, size)
X = _N.require(X, requirements = 'C')
niter = 10
# Contiguous
print "Running optimized with ctypes"
for i in range(niter):
at, et, kt = lpc2(X, order, axis = 1)
a, e, k = _lpc2_py(X, order, axis = 1)
_N.testing.assert_array_almost_equal(a, at, 10)
_N.testing.assert_array_almost_equal(e, et, 10)
_N.testing.assert_array_almost_equal(k, kt, 10)
## Non contiguous
#print "Running optimized with ctypes (non contiguous)"
#def ncontig(*args, **kargs):
# return lpc2(*args, **kargs)
#X = _N.require(X, requirements = 'F')
#for i in range(niter):
# at, et, kt = ncontig(X, order, axis = 1)
#a, e, k = _lpc2_py(X, order, axis = 1)
#_N.testing.assert_array_almost_equal(a, at, 10)
#_N.testing.assert_array_almost_equal(e, et, 10)
#_N.testing.assert_array_almost_equal(k, kt, 10)
print "Benchmark func done"
if __name__ == '__main__':
import hotshot, hotshot.stats
profile_file = 'lpc.prof'
prof = hotshot.Profile(profile_file, lineevents=1)
prof.runcall(bench)
p = hotshot.stats.load(profile_file)
print p.sort_stats('cumulative').print_stats(20)
prof.close()
|
