def setUp(self):
self.samplerate = SR44100()
rs = resampled(resample_to=self.samplerate)
wscheme = HalfLapped()
@simple_in_memory_settings
class Document(rs):
windowed = ArrayWithUnitsFeature(
SlidingWindow,
wscheme=wscheme,
wfunc=OggVorbisWindowingFunc(),
needs=rs.resampled,
store=False)
fft = ArrayWithUnitsFeature(
FFT,
needs=windowed,
store=False)
centroid = ArrayWithUnitsFeature(
SpectralCentroid,
needs=fft,
store=True)
ss = SineSynthesizer(self.samplerate)
chunks = \
[ss.synthesize(Seconds(1), [440 * i]) for i in range(1, 6)]
self.audio = \
AudioSamples(ArrayWithUnits.concat(chunks), self.samplerate)
_id = Document.process(meta=self.audio.encode())
self.doc = Document(_id)
def setUp(self):
self.samplerate = SR22050()
rs = resampled(resample_to=self.samplerate)
window_size = Picoseconds(int(1e12))
wscheme = SampleRate(window_size, window_size)
@simple_in_memory_settings
class Document(rs):
windowed = ArrayWithUnitsFeature(
SlidingWindow,
wscheme=wscheme,
needs=rs.resampled,
store=False)
dct = ArrayWithUnitsFeature(
DCTIV,
needs=windowed,
store=True)
ss = SineSynthesizer(self.samplerate)
self.audio = ss.synthesize(Seconds(5), [440., 660., 880.])
_id = Document.process(meta=self.audio.encode())
self.doc = Document(_id)
def test_can_apply_empty_time_slice_to_wrapper(self):
synth = SineSynthesizer(SR11025())
samples = synth.synthesize(Seconds(10))
encoded = samples.encode(fmt='OGG', subtype='VORBIS')
wrapper = OggVorbisWrapper(encoded)
samples = wrapper[TimeSlice()]
expected = Seconds(10) / Seconds(1)
actual = samples.end / Seconds(1)
self.assertAlmostEqual(expected, actual, places=6)
def test_can_convert_to_categorical_distribution(self):
samplerate = SR11025()
synth = SineSynthesizer(samplerate)
samples = synth.synthesize(Seconds(4), [220, 440, 880])
_, windowed = samples.sliding_window_with_leftovers(
TimeSlice(duration=samplerate.frequency * 512),
TimeSlice(duration=samplerate.frequency * 256))
c = categorical(windowed, mu=255)
self.assertEqual(windowed.shape + (255 + 1,), c.shape)
np.testing.assert_allclose(c.sum(axis=-1), 1)
def test_can_convert_to_categorical_distribution(self):
samplerate = SR11025()
synth = SineSynthesizer(samplerate)
samples = synth.synthesize(Seconds(4), [220, 440, 880])
_, windowed = samples.sliding_window_with_leftovers(
TimeSlice(duration=samplerate.frequency * 512),
TimeSlice(duration=samplerate.frequency * 256))
c = categorical(windowed, mu=255)
self.assertEqual(windowed.shape + (255 + 1, ), c.shape)
np.testing.assert_allclose(c.sum(axis=-1), 1)
def test_can_invert_categorical_distribution(self):
samplerate = SR11025()
synth = SineSynthesizer(samplerate)
samples = synth.synthesize(Seconds(4), [220, 440, 880])
_, windowed = samples.sliding_window_with_leftovers(
TimeSlice(duration=samplerate.frequency * 512),
TimeSlice(duration=samplerate.frequency * 256))
c = categorical(windowed, mu=255)
inverted = inverse_categorical(c, mu=255)
self.assertEqual(windowed.shape, inverted.shape)
self.assertIsInstance(inverted, ArrayWithUnits)
self.assertSequenceEqual(windowed.dimensions, inverted.dimensions)
def test_can_invert_categorical_distribution(self):
samplerate = SR11025()
synth = SineSynthesizer(samplerate)
samples = synth.synthesize(Seconds(4), [220, 440, 880])
_, windowed = samples.sliding_window_with_leftovers(
TimeSlice(duration=samplerate.frequency * 512),
TimeSlice(duration=samplerate.frequency * 256))
c = categorical(windowed, mu=255)
inverted = inverse_categorical(c, mu=255)
self.assertEqual(windowed.shape, inverted.shape)
self.assertIsInstance(inverted, ArrayWithUnits)
self.assertSequenceEqual(windowed.dimensions, inverted.dimensions)
def test_perfect_reconstruction(self):
synth = SineSynthesizer(SR22050())
audio = synth.synthesize(Seconds(1), [440., 660., 880.])
node = DCTIV()
coeffs = node._process_raw(audio[None, :])
recon = node._process_raw(coeffs)[0]
# see? it's do-able w/ dct
dct_coeffs = scipy.fftpack.dct(audio[None, :], norm='ortho')
inverse_dct = scipy.fftpack.idct(dct_coeffs, norm='ortho')[0]
np.testing.assert_almost_equal(inverse_dct, audio, decimal=4)
np.testing.assert_almost_equal(recon, audio, decimal=4)
def test_perfect_reconstruction_using_overlap_add(self):
synth = SineSynthesizer(SR22050())
audio = synth.synthesize(Seconds(10), [440., 660., 880.])
sr = SampleRate(duration=Seconds(1), frequency=Milliseconds(500))
windowed = audio.sliding_window(sr)
mdct = MDCT()
coeffs = list(mdct._process(windowed * OggVorbisWindowingFunc()))[0]
mdct_synth = MDCTSynthesizer()
recon = mdct_synth.synthesize(coeffs)
# take a slice, so we can ignore boundary conditions
slce = TimeSlice(start=Seconds(1), duration=Seconds(8))
np.testing.assert_allclose(recon[slce], audio[slce])
def setUp(self):
self.samplerate = SR22050()
rs = resampled(resample_to=self.samplerate)
wscheme = HalfLapped()
@simple_in_memory_settings
class Document(rs):
windowed = ArrayWithUnitsFeature(
SlidingWindow,
wscheme=wscheme,
wfunc=OggVorbisWindowingFunc(),
needs=rs.resampled,
store=False)
fft = ArrayWithUnitsFeature(
FFT,
needs=windowed,
store=False)
flatness = ArrayWithUnitsFeature(
SpectralFlatness,
needs=fft,
store=True)
# create a pure sine wave that fades out
ss = SineSynthesizer(self.samplerate)
sine = ss.synthesize(Seconds(5), [440.])
sine_envelope = np.linspace(1.0, 0.0, len(sine))
sine *= sine_envelope
# create noise
ns = NoiseSynthesizer(self.samplerate)
noise = ns.synthesize(Seconds(5))
noise_envelope = np.linspace(0.0, 1.0, len(noise))
noise *= noise_envelope
# mix the sine wave and noise together
self.audio = sine + noise
_id = Document.process(meta=self.audio.encode())
self.doc = Document(_id)
def setUp(self):
self.samplerate = SR11025()
rs = resampled(resample_to=self.samplerate)
wscheme = HalfLapped()
@simple_in_memory_settings
class Document(rs):
windowed = ArrayWithUnitsFeature(SlidingWindow,
wscheme=wscheme,
needs=rs.resampled,
store=False)
mdct = ArrayWithUnitsFeature(MDCT, needs=windowed, store=True)
ss = SineSynthesizer(self.samplerate)
self.audio = ss.synthesize(Seconds(5), [440., 660., 880.])
_id = Document.process(meta=self.audio.encode())
self.doc = Document(_id)
def setUp(self):
self.samplerate = SR11025()
rs = resampled(resample_to=self.samplerate)
wscheme = HalfLapped()
@simple_in_memory_settings
class Document(rs):
windowed = ArrayWithUnitsFeature(
SlidingWindow,
wscheme=wscheme,
needs=rs.resampled,
store=False)
mdct = ArrayWithUnitsFeature(
MDCT,
needs=windowed,
store=True)
ss = SineSynthesizer(self.samplerate)
self.audio = ss.synthesize(Seconds(5), [440., 660., 880.])
_id = Document.process(meta=self.audio.encode())
self.doc = Document(_id)
def test_pad_with_samples_adds_silence_at_end(self):
synth = SineSynthesizer(SR11025())
samples = synth.synthesize(Seconds(2))
padded = samples.pad_with_silence(Seconds(4))
silence = padded[TimeSlice(start=Seconds(2))]
self.assertEqual(0, silence.sum())
def test_pad_with_samples_adds_silence_at_end(self):
synth = SineSynthesizer(SR11025())
samples = synth.synthesize(Seconds(2))
padded = samples.pad_with_silence(Seconds(4))
silence = padded[TimeSlice(start=Seconds(2))]
self.assertEqual(0, silence.sum())
