+from numpy.testing import TestCase, run_module_suite
+from numpy.testing import assert_equal, assert_almost_equal
+# WARNING: numpy also has an fft object
+from _aubio import cvec, specdesc
+from numpy import array, shape, arange, zeros, log
+from math import pi
+
+class aubio_specdesc(TestCase):
+
+ def test_members(self):
+ o = specdesc()
+ assert_equal ([o.buf_size, o.channels, o.method],
+ [1024, 1, "default"])
+ o = specdesc("complex", 512, 2)
+ assert_equal ([o.buf_size, o.channels, o.method],
+ [512, 2, "complex"])
+
+ def test_hfc(self):
+ o = specdesc("hfc")
+ c = cvec()
+ assert_equal( 0., o(c))
+ a = arange(c.length, dtype='float32')
+ c.norm = a
+ assert_equal (a, c.norm[0])
+ assert_equal ( sum(a*(a+1)), o(c))
+
+ def test_complex(self):
+ o = specdesc("complex")
+ c = cvec()
+ assert_equal( 0., o(c))
+ a = arange(c.length, dtype='float32')
+ c.norm = a
+ assert_equal (a, c.norm[0])
+ # the previous run was on zeros, so previous frames are still 0
+ # so we have sqrt ( abs ( r2 ^ 2) ) == r2
+ assert_equal ( sum(a), o(c))
+ # second time. c.norm = a, so, r1 = r2, and the euclidian distance is 0
+ assert_equal ( 0, o(c))
+
+ def test_phase(self):
+ o = specdesc("phase")
+ c = cvec()
+ assert_equal( 0., o(c))
+
+ def test_kl(self):
+ o = specdesc("kl")
+ c = cvec()
+ assert_equal( 0., o(c))
+ a = arange(c.length, dtype='float32')
+ c.norm = a
+ assert_almost_equal( sum(a * log(1.+ a/1.e-10 ) ) / o(c), 1., decimal=6)
+
+ def test_mkl(self):
+ o = specdesc("mkl")
+ c = cvec()
+ assert_equal( 0., o(c))
+ a = arange(c.length, dtype='float32')
+ c.norm = a
+ assert_almost_equal( sum(log(1.+ a/1.e-10 ) ) / o(c), 1, decimal=6)
+
+ def test_specflux(self):
+ o = specdesc("specflux")
+ c = cvec()
+ assert_equal( 0., o(c))
+ a = arange(c.length, dtype='float32')
+ c.norm = a
+ assert_equal( sum(a), o(c))
+ assert_equal( 0, o(c))
+ c.norm = zeros(c.length, dtype='float32')
+ assert_equal( 0, o(c))
+
+ def test_centroid(self):
+ o = specdesc("centroid")
+ c = cvec()
+ # make sure centroid of zeros is zero
+ assert_equal( 0., o(c))
+ a = arange(c.length, dtype='float32')
+ c.norm = a
+ centroid = sum(a*a) / sum(a)
+ assert_almost_equal (centroid, o(c), decimal = 2)
+
+ c.norm = a * .5
+ assert_almost_equal (centroid, o(c), decimal = 2)
+
+ def test_spread(self):
+ o = specdesc("spread")
+ c = cvec()
+ assert_equal( 0., o(c))
+ a = arange(c.length, dtype='float32')
+ c.norm = a
+ centroid = sum(a*a) / sum(a)
+ spread = sum( (a - centroid)**2 *a) / sum(a)
+ assert_almost_equal (spread, o(c), decimal = 2)
+
+ c.norm = a * 3
+ assert_almost_equal (spread, o(c), decimal = 2)
+
+ def test_skewness(self):
+ o = specdesc("skewness")
+ c = cvec()
+ assert_equal( 0., o(c))
+ a = arange(c.length, dtype='float32')
+ c.norm = a
+ centroid = sum(a*a) / sum(a)
+ spread = sum( (a - centroid)**2 *a) / sum(a)
+ skewness = sum( (a - centroid)**3 *a) / sum(a) / spread **1.5
+ assert_almost_equal (skewness, o(c), decimal = 2)
+
+ c.norm = a * 3
+ assert_almost_equal (skewness, o(c), decimal = 2)
+
+ def test_kurtosis(self):
+ o = specdesc("kurtosis")
+ c = cvec()
+ assert_equal( 0., o(c))
+ a = arange(c.length, dtype='float32')
+ c.norm = a
+ centroid = sum(a*a) / sum(a)
+ spread = sum( (a - centroid)**2 *a) / sum(a)
+ kurtosis = sum( (a - centroid)**4 *a) / sum(a) / spread **2
+ assert_almost_equal (kurtosis, o(c), decimal = 2)
+
+ def test_slope(self):
+ o = specdesc("slope")
+ c = cvec()
+ assert_equal( 0., o(c))
+ a = arange(c.length * 2, 0, -2, dtype='float32')
+ k = arange(c.length, dtype='float32')
+ c.norm = a
+ num = len(a) * sum(k*a) - sum(k)*sum(a)
+ den = (len(a) * sum(k**2) - sum(k)**2)
+ slope = num/den/sum(a)
+ assert_almost_equal (slope, o(c), decimal = 5)
+
+ a = arange(0, c.length * 2, +2, dtype='float32')
+ c.norm = a
+ num = len(a) * sum(k*a) - sum(k)*sum(a)
+ den = (len(a) * sum(k**2) - sum(k)**2)
+ slope = num/den/sum(a)
+ assert_almost_equal (slope, o(c), decimal = 5)
+
+ a = arange(0, c.length * 2, +2, dtype='float32')
+ c.norm = a * 2
+ assert_almost_equal (slope, o(c), decimal = 5)
+
+ def test_decrease(self):
+ o = specdesc("decrease")
+ c = cvec()
+ assert_equal( 0., o(c))
+ a = arange(c.length * 2, 0, -2, dtype='float32')
+ k = arange(c.length, dtype='float32')
+ c.norm = a
+ decrease = sum((a[1:] - a [0]) / k[1:]) / sum(a[1:])
+ assert_almost_equal (decrease, o(c), decimal = 5)
+
+ a = arange(0, c.length * 2, +2, dtype='float32')
+ c.norm = a
+ decrease = sum((a[1:] - a [0]) / k[1:]) / sum(a[1:])
+ assert_almost_equal (decrease, o(c), decimal = 5)
+
+ a = arange(0, c.length * 2, +2, dtype='float32')
+ c.norm = a * 2
+ decrease = sum((a[1:] - a [0]) / k[1:]) / sum(a[1:])
+ assert_almost_equal (decrease, o(c), decimal = 5)
+
+ def test_rolloff(self):
+ o = specdesc("rolloff")
+ c = cvec()
+ assert_equal( 0., o(c))
+ a = arange(c.length * 2, 0, -2, dtype='float32')
+ k = arange(c.length, dtype='float32')
+ c.norm = a
+ cumsum = .95*sum(a*a)
+ i = 0; rollsum = 0
+ while rollsum < cumsum:
+ rollsum += a[i]*a[i]
+ i+=1
+ rolloff = i
+ assert_equal (rolloff, o(c))
+
+if __name__ == '__main__':
+ from unittest import main
+ main()