class ShortTimeFourierTransform(BaseModel):
meta = JSONFeature(MetaData, store=True, encoder=AudioMetaDataEncoder)
raw = ByteStreamFeature(ByteStream,
chunksize=chunksize_bytes,
needs=meta,
store=False)
ogg = OggVorbisFeature(OggVorbis, needs=raw, store=True)
pcm = AudioSamplesFeature(AudioStream, needs=raw, store=False)
resampled = AudioSamplesFeature(Resampler,
needs=pcm,
samplerate=resample_to,
store=store_resampled)
windowed = ArrayWithUnitsFeature(SlidingWindow,
needs=resampled,
wscheme=wscheme,
wfunc=OggVorbisWindowingFunc(),
store=store_windowed)
fft = ArrayWithUnitsFeature(FFT,
padding_samples=fft_padding_samples,
needs=windowed,
store=store_fft)
def test_square_form_no_overlap_add(self):
samplerate = SR11025()
BaseModel = stft(resample_to=samplerate)
windowing_func = OggVorbisWindowingFunc()
scale = GeometricScale(20, 5000, 0.1, 25)
@simple_in_memory_settings
class Document(BaseModel):
long_windowed = ArrayWithUnitsFeature(
SlidingWindow,
wscheme=SampleRate(frequency=Milliseconds(500),
duration=Seconds(1)),
wfunc=windowing_func,
needs=BaseModel.resampled,
store=True)
dct = ArrayWithUnitsFeature(DCT,
scale_always_even=True,
needs=long_windowed,
store=True)
mdct = FrequencyAdaptiveFeature(FrequencyAdaptiveTransform,
transform=scipy.fftpack.idct,
scale=scale,
needs=dct,
store=True)
synth = TickSynthesizer(SR22050())
samples = synth.synthesize(Seconds(5), Milliseconds(200))
_id = Document.process(meta=samples.encode())
doc = Document(_id)
square = doc.mdct.square(30)
self.assertEqual(3, square.ndim)
self.assertEqual(30, square.shape[1])
self.assertEqual(25, square.shape[2])
class Document(rs):
long_windowed = ArrayWithUnitsFeature(
SlidingWindow,
wscheme=SampleRate(Milliseconds(500), Seconds(1)),
wfunc=OggVorbisWindowingFunc(),
needs=rs.resampled,
store=True)
long_fft = ArrayWithUnitsFeature(FFT,
needs=long_windowed,
store=True)
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)
class FrequencyAdaptive(BaseModel):
long_windowed = ArrayWithUnitsFeature(SlidingWindow,
wscheme=long_window_sample_rate,
wfunc=OggVorbisWindowingFunc(),
needs=BaseModel.resampled,
store=False)
long_fft = ArrayWithUnitsFeature(FFT, needs=long_windowed, store=False)
freq_adaptive = FrequencyAdaptiveFeature(
FrequencyAdaptiveTransform,
transform=np.fft.irfft,
scale=scale,
check_scale_overlap_ratio=check_scale_overlap_ratio,
window_func=np.hanning,
needs=long_fft,
store=store_freq_adaptive)
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])
class Document(BaseModel):
raw = ByteStreamFeature(ByteStream,
chunksize=2 * 44100 * 30 * 2,
store=True)
ogg = OggVorbisFeature(OggVorbis, needs=raw, store=True)
pcm = AudioSamplesFeature(AudioStream, needs=raw, store=True)
resampled = AudioSamplesFeature(Resampler,
needs=pcm,
samplerate=samplerate,
store=True)
windowed = ArrayWithUnitsFeature(SlidingWindow,
needs=resampled,
wscheme=windowing_scheme,
wfunc=OggVorbisWindowingFunc(),
store=False)
fft = ArrayWithUnitsFeature(FFT, needs=windowed, store=True)
chroma = ArrayWithUnitsFeature(Chroma,
needs=fft,
frequency_band=band,
store=True)
bark = ArrayWithUnitsFeature(BarkBands,
needs=fft,
frequency_band=band,
store=True)
bfcc = ArrayWithUnitsFeature(BFCC, needs=bark, store=True)
bfcc_sliding_window = ArrayWithUnitsFeature(
SlidingWindow,
needs=bfcc,
wscheme=windowing_scheme * Stride(frequency=2, duration=4),
store=True)
bfcc_pooled = ArrayWithUnitsFeature(Max,
needs=bfcc_sliding_window,
axis=1,
store=True)
class AudioGraph(BaseModel):
meta = JSONFeature(MetaData, store=True, encoder=AudioMetaDataEncoder)
raw = ByteStreamFeature(ByteStream,
chunksize=chunksize_bytes,
needs=meta,
store=False)
ogg = OggVorbisFeature(OggVorbis, needs=raw, store=True)
pcm = AudioSamplesFeature(AudioStream, needs=raw, store=False)
resampled = AudioSamplesFeature(Resampler,
needs=pcm,
samplerate=resample_to,
store=False)
windowed = ArrayWithUnitsFeature(SlidingWindow,
needs=resampled,
wscheme=HalfLapped(),
wfunc=OggVorbisWindowingFunc(),
store=False)
dct = ArrayWithUnitsFeature(DCT, needs=windowed, store=True)
fft = ArrayWithUnitsFeature(FFT, needs=windowed, store=store_fft)
bark = ArrayWithUnitsFeature(BarkBands,
needs=fft,
frequency_band=band,
store=True)
centroid = ArrayWithUnitsFeature(SpectralCentroid,
needs=bark,
store=True)
chroma = ArrayWithUnitsFeature(Chroma,
needs=fft,
frequency_band=band,
store=True)
bfcc = ArrayWithUnitsFeature(BFCC, needs=fft, store=True)