def feature_vector(base_name):
print base_name
f = sound_path(base_name)
f, sr = librosa.load(f)
fv = np.array([])
fv = np.append(fv, calculateRMSE(f))
fv = np.append(fv, compression_rate(base_name))
fv = np.append(fv, ExtractTemporalSparcity(f))
fv = np.append(fv, ExtractMelSpectraSparcityFeatures(f))
fv = np.append(fv, ExtractCQSpectraSparcityFeatures(f))
fv = np.append(fv, ExtractSTFTSpectraSparcityFeatures(f))
fv = np.append(fv, calculateRMSETimeHomogeneity(f))
# for moment in moments:
# fv = np.append(fv, calculateSpectraStatisticTimeHomogeneity(f, librosa.cqt, moment, 10))
fv = np.append(fv, calculateCrossCorrelations(f, spectra[0]))
# fv = np.append(fv, twoLayerTransform(f, spectra[0]))
# fv = np.append(fv, calculateModulationSubbandKStatisticTimeHomogineity(f, spectra[0], np.mean, 10))
# fv = np.append(fv, calculateModulationSubbandKStatisticTimeHomogineity(f, spectra[0], np.var, 10))
"""
for spectrum in spectra:
for moment in moments:
fv = np.append(fv, calculateSpectraStatisticTimeHomogeneity(f, spectrum, moment, 10))
"""
return fv
def compression_rate(base_name):
uncompressed_size = os.path.getsize(sound_path(base_name))
compressed_size = os.path.getsize(compressed_path(base_name))
assert compressed_size > 0
return float(uncompressed_size)/compressed_size
