def test_extractor_without_sink():
ex = Extractor(A())
sc = ArraySource(list(range(10)), blocksize=1, overlap=0, samplerate=1)
for i, res in enumerate(ex.extract(sc)):
assert i == res['A']
ex.reset()
assert ex.featureset['A']._started is False
def test_array_source():
a = ArraySource(list(range(10)), blocksize=2, overlap=1, samplerate=1)
last = 0
for i, b in enumerate(a):
data, index = b
assert len(b) == 2
assert last == data[0]
last = data[-1]
def test_extractor_with_sink():
ex = Extractor(A(), A(name='hidden_a').hide())
sc = ArraySource(list(range(10)), blocksize=1, overlap=0, samplerate=1)
sk = DefaultDictSink()
ex.extract(sc, sk)
assert list(sk['results']['A']) == list(range(10))
assert any(sk['hiddenfeatures'])
assert any(sk['features'])
def test_windowed_signal():
block = np.ones(10)
sc = ArraySource(block, samplerate=1)
wsf = WindowedSignal(window='hanning', size=10)
wsf.on_start(sc)
res = wsf.process((block, 0), {})
assert np.allclose(res, wsf.w)
block = np.ones((10, 2))
sc = ArraySource(block, samplerate=1, blocksize=10)
wsf = WindowedSignal(window='blackman')
wsf.on_start(sc)
res = wsf.process((block, 0), {})
assert np.allclose(res, wsf.w)
sc = ArraySource(block, samplerate=1, blocksize=10)
wsf = WindowedSignal(window='rect')
wsf.on_start(sc)
res = wsf.process((block, 0), {})
assert np.allclose(res, wsf.w)
def test_abstract_preprocess():
with pytest.raises(TypeError):
Preprocess()
class P(Preprocess):
def process(self, data):
return data
src = ArraySource(np.ones((10, 2)), samplerate=10)
pp = P(src)
list(pp)
def test_spectral_features_no_window_branch():
from sigfeat.feature.spectral import AbsRfft
features = [
AbsRfft(window=False),
SpectralCentroid(),
SpectralFlatness(),
SpectralFlux(),
SpectralCrestFactor(),
SpectralRolloff(),
]
etr = Extractor(*features)
x = np.sin(1000 * 2 * np.pi * np.linspace(0, 1, 44100))
src = ArraySource(np.tile(x, (2, 1)).T,
samplerate=44100,
blocksize=2048,
overlap=1024)
etr.extract(src, DefaultDictSink())
def test_delta():
class A(Feature):
def process(self, data, res):
return float(data[0])
ex = Extractor(
Delta(A()),
Delta(A(), order=2)
)
x = np.arange(0.0, 10.0, 2)
src = ArraySource(x, samplerate=1, blocksize=1, channels=1)
snk = ex.extract(src, DefaultDictSink())
res = snk['results']
assert np.allclose(np.array(res['A']).flatten(), x)
assert np.mean(np.array(res['dA']).flatten()[1:]) == 2.0
assert np.mean(np.array(res['ddA']).flatten()[2:]) == 0.0
assert res['dA'][0] is None
assert res['ddA'][0] is None
def test_spectral_features():
features = [
SpectralCentroid(),
SpectralFlatness(),
SpectralFlux(),
SpectralCrestFactor(),
SpectralRolloff(),
SpectralSpread(),
SpectralSkewness(),
SpectralKurtosis(),
SpectralSlope(),
]
etr = Extractor(*features)
x = np.sin(1000 * 2 * np.pi * np.linspace(0, 1, 44100))
src = ArraySource(x, samplerate=44100, blocksize=2048, overlap=1024)
snk = etr.extract(src, DefaultDictSink())
res = snk['results']
assert abs(1 - np.median(res['SpectralCentroid']) / 1000) < 1e-3
assert abs(np.median(res['SpectralFlatness'])) < 1e-3
def test_temporal_features():
features = [
CrestFactor(),
ZeroCrossingRate(),
StatMoments(),
CentroidAbsSignal(),
FlatnessAbsSignal(),
RootMeanSquare(),
Peak(),
Kurtosis(),
Skewness(),
StandardDeviation(),
]
ex = Extractor(*features)
x = np.sin(1000*2*np.pi*np.linspace(0, 1, 44100))
src = ArraySource(x, samplerate=44100, blocksize=4096)
snk = ex.extract(src, DefaultDictSink())
res = snk['results']
assert abs(np.median(res['CrestFactor'])-2**0.5) < 1e-3
assert abs(1-np.median(res['ZeroCrossingRate'])/2000.0) < 1e-2
assert abs(2**-0.5-np.median(res['RootMeanSquare'])) < 1e-3
assert abs(1-np.median(res['Peak'])) < 1e-4
from sigfeat import feature as fts
t = np.linspace(0, 2, 2*44100)
# x = np.sin(2*np.pi*1000*t)
x = chirp(
t,
f0=1000,
t1=2,
f1=4000,
method='log'
)
src = ArraySource(
np.tile(x, (2, 1)).T,
samplerate=44100,
blocksize=2048,
overlap=1024)
features = (
fts.Index(),
fts.RootMeanSquare(),
fts.Peak(),
fts.CrestFactor(),
fts.ZeroCrossingRate(),
fts.SpectralFlux(),
fts.SpectralCentroid(),
fts.SpectralFlatness(),
fts.SpectralCrestFactor(),
)
def mkas():
return ArraySource(np.random.randn(44100),
samplerate=44100,
blocksize=2048,
overlap=1024)
def test_sum_mix():
src = ArraySource(np.ones((10, 2)), samplerate=10, blocksize=1)
pp = SumMix(src)
for data in pp:
assert data[0] == 2.0
def test_mean_mix():
src = ArraySource(np.ones((10, 2)), channels=2, samplerate=10, blocksize=1)
pp = MeanMix(src)
for data in pp:
assert data[0] == 1.0