def test_againstMIR_testalt_sample_both(self):
val = extractor.zcr(test_alt,
p='sample',
d='both',
decomposition=False)
MIRVAL = 0.0243
assert np.abs(val - MIRVAL) <= percentError * MIRVAL
def test_againstLIBROSA_beethoven(self):
my_val = extractor.zcr(beet, win_length=n_fft / sr, decomposition=True)
lib_val = librosa.feature.zero_crossing_rate(y=beet,
frame_length=n_fft,
hop_length=n_fft / 2)
corr = calculateZcorr(my_val, retrieveLibrosaValue(lib_val))
assert corr >= 0.95 # assert 95% correlation b/w zscores
def test_againstMIR_beethoven_sample_both(self):
val = extractor.zcr(beet,
sr,
p='sample',
d='both',
decomposition=False)
MIRVAL = 0.028678
assert np.abs(val - MIRVAL) <= percentError * MIRVAL
def zcr(self, use_librosa=False, decomposition=True, p='second', d='one',
hop_length=None, n_fft=None):
"""
Get the zero crossing rate feature.
:parameters:
- use_librosa : boolean. Whether to use librosa or extractor
feature extraction function. Default False.
- decomposition : boolean. Whether to look at a time-series or
an overall root mean square analysis of the piece.
- p : boolean. p='second' or p='sample'. Calculate zero crossing
rate per these two units of time. Default is 'second'.
- d : boolean. d='one' or d='both'. Calculate zero crossing rate
from one direction or both directions of approach to x-axis.
Default is 'one'
- n_fft : integer. The window length [samples].
Default 2048 [samples].
- hop_length : float. Hop length (i.e. overlap) between the
frames in the time series analysis [samples].
Default is half-overlap.
- n_fft : integer. The window length [samples]. For using with
librosa. Default 2048 [samples].
"""
if n_fft is None:
n_fft = self.n_fft
if hop_length is None:
hop_length = int(n_fft / 2)
if use_librosa:
tmp = librosa.feature.zero_crossing_rate(self.audio,
n_fft=n_fft, hop_length=hop_length)
self.insertFeature(tmp, 'ZCR', hop_length, full=False)
self.insertExtractionParameters('ZCR',
dict(n_fft=n_fft, hop_length=hop_length,
librosa=use_librosa, decomposition=True))
else:
tmp = extractor.zcr(self.audio, sr=self.sr, p=p,
d=d, n_fft=n_fft, hop_length=hop_length,
pad=self.pad, decomposition=decomposition)
self.insertFeature(tmp, 'ZCR', hop_length, full=not decomposition)
self.insertExtractionParameters('ZCR',
dict(hop_length=hop_length,
decomposition=decomposition,
n_fft=n_fft, per=p, direction=d,
pad=self.pad, librosa=use_librosa))
return tmp
def test_againstMIR_testalt(self):
val = extractor.zcr(test_alt, decomposition=False)
MIRVAL = 536
assert np.abs(val - MIRVAL) <= percentError * MIRVAL
def test_againstMIR_beethoven(self):
val = extractor.zcr(beet, decomposition=False)
MIRVAL = 632.3395
# within percent error of the MIR value
assert np.abs(val - MIRVAL) <= percentError * MIRVAL
def test_ones_and_minusones_sample_one(self):
a = np.ones(10)
a[1::2] = -1
npt.assert_equal(extractor.zcr(a, p='sample', decomposition=False),
0.9 / 2)
def test_ones_and_minusones_second_one(self):
a = np.ones(10)
a[1::2] = -1
sr = 22050
npt.assert_equal(extractor.zcr(a, sr=sr, decomposition=False),
0.9 * sr / 2)
def test_ones_and_minusones_second_both(self):
a = np.ones(10)
a[1::2] = -1
npt.assert_equal(extractor.zcr(a, sr=2, d='both', decomposition=False),
0.9 * 2)
def test_ones(self):
a = np.ones(10)
npt.assert_equal(extractor.zcr(a, decomposition=False), 0.0)
