def getpeaks(self, sample):
from numarray import nonzero
from numarray.fft import real_fft
fft = abs(real_fft(sample))
# We compare each point to the point on either side - we must be
# greater than both.
# To do this, we left-shift and right-shift the array by one to
# compare.
peaks = nonzero((fft[1:-1] > fft[:-2]) * (fft[1:-1] > fft[2:]))
if not len(peaks) or not len(peaks[0]):
return ()
# Avoiding 'zip; sort; reverse' will give a big speedup.
try:
peakvals = zip(fft[1:-1].take(peaks)[0], peaks[0])
except:
print("pppp", peaks)
raise
peakvals.sort(); peakvals.reverse()
if len(peakvals) > 2:
(val1,ind1),(val2,ind2),(val3,ind3) = peakvals[:3]
# Peak 3 should be no more than 2/3 the size of peak 2
# For GSM/Speex mangled audio, peak3 is about 0.55 * peak2
# at the worst case.
if val2 > 1.5*val3:
# Add one to the index because we took [1:-1] before
return (ind2+1,ind1+1)
else:
return ()
elif len(peakvals) == 2:
(val2,ind2),(val1,ind1) = peakvals
# Add one to the index because we took [1:-1] before
return (ind2+1,ind1+1)
else:
(val1,ind1), = peakvals
return (ind1,0)
def getpeaks(self, sample):
from numarray import nonzero
from numarray.fft import real_fft
fft = abs(real_fft(sample))
# We compare each point to the point on either side - we must be
# greater than both.
# To do this, we left-shift and right-shift the array by one to
# compare.
peaks = nonzero((fft[1:-1] > fft[:-2]) * (fft[1:-1] > fft[2:]))
if not len(peaks) or not len(peaks[0]):
return ()
# Avoiding 'zip; sort; reverse' will give a big speedup.
try:
peakvals = zip(fft[1:-1].take(peaks)[0], peaks[0])
except:
print "pppp", peaks
raise
peakvals.sort(); peakvals.reverse()
if len(peakvals) > 2:
(val1,ind1),(val2,ind2),(val3,ind3) = peakvals[:3]
# Peak 3 should be no more than 2/3 the size of peak 2
# For GSM/Speex mangled audio, peak3 is about 0.55 * peak2
# at the worst case.
if val2 > 1.5*val3:
# Add one to the index because we took [1:-1] before
return (ind2+1,ind1+1)
else:
return ()
elif len(peakvals) == 2:
(val2,ind2),(val1,ind1) = peakvals
# Add one to the index because we took [1:-1] before
return (ind2+1,ind1+1)
else:
(val1,ind1), = peakvals
return (ind1,0)
def multispec(data, doxs=1):
"""Compute power and cross spectra.
power, cross = multispec(data[, doxs=1])
data is expected to be a numeric sequence, representing a binned light
curve.
If doxs is false or data is one-dimensional, the cross spectra will not
be calculated as described below, and only the power spectra will be
returned.
If data is one-dimensional, its periodogram will be calculated.
If data is two-dimensional, it will be interpreted as an array of
one-dimensional curves. The periodogram will be calculated for each,
and the cross spectra of all but the first relative to the first.
If data has more than two dimensions, it will be interpreted as an array
of two-dimensional sets, and each will be treated as described above."""
# Figure out what we've got.
data = num.asarray(data)
npoints = data.shape[-1]
if len(data.shape) < 2:
doxs = 0
# crunch
transform = None
if num.__name__ == 'numarray':
transform = FFT.real_fft(data)
else:
transform = FFT.fft(data)
conj = num.conjugate(transform)
# Get the periodogram.
power = transform * conj
power = num.array(power.real)
if doxs:
# Calculate the cross spectrum.
cross = transform[..., :1, :] * conj[..., 1:, :]
return power, cross
else:
return power
def multispec(data, doxs=1):
"""Compute power and cross spectra.
power, cross = multispec(data[, doxs=1])
data is expected to be a numeric sequence, representing a binned light
curve.
If doxs is false or data is one-dimensional, the cross spectra will not
be calculated as described below, and only the power spectra will be
returned.
If data is one-dimensional, its periodogram will be calculated.
If data is two-dimensional, it will be interpreted as an array of
one-dimensional curves. The periodogram will be calculated for each,
and the cross spectra of all but the first relative to the first.
If data has more than two dimensions, it will be interpreted as an array
of two-dimensional sets, and each will be treated as described above."""
# Figure out what we've got.
data = num.asarray(data)
npoints = data.shape[-1]
if len(data.shape) < 2:
doxs = 0
# crunch
transform = None
if num.__name__ == 'numarray':
transform = FFT.real_fft(data)
else:
transform = FFT.fft(data)
conj = num.conjugate(transform)
# Get the periodogram.
power = transform * conj
power = num.array(power.real)
if doxs:
# Calculate the cross spectrum.
cross = transform[..., :1, :] * conj[..., 1:, :]
return power, cross
else:
return power
