def spectral_entropy(X_dataA,
X_dataB,
X_testdataA,
y_labelA,
y_labelB,
if_plot=False):
"""Get the Spectral entropy"""
entropy_freqA = normalize(
np.array([
audioFeatureExtraction.stSpectralEntropy(X, numOfShortBlocks=150)
for X in X_dataA
]))
entropy_freqB = normalize(
np.array([
audioFeatureExtraction.stSpectralEntropy(X, numOfShortBlocks=50)
for X in X_dataB
]))
entr_freqtestA = normalize(
np.array([
audioFeatureExtraction.stSpectralEntropy(X, numOfShortBlocks=150)
for X in X_testdataA
]))
if if_plot:
featurePlot(entropy_freqA,
'A',
y_labelA,
title='Spectral Entropy (set A) vs Feature Labels (set A)')
featurePlot(entropy_freqB,
'B',
y_labelB,
title='Spectral Entropy (set B) vs Feature Labels (set B)')
return entropy_freqA, entropy_freqB, entr_freqtestA
def signal_energy_frame(x_dataA,
x_dataB,
x_testdataA,
y_labelA,
y_labelB,
if_plot=False):
"""signal energy of frame"""
energyA = normalize(
np.array([audioFeatureExtraction.stEnergy(x) for x in x_dataA]))
energyB = normalize(
np.array([audioFeatureExtraction.stEnergy(x) for x in x_dataB]))
ener_testA = normalize(
np.array([audioFeatureExtraction.stEnergy(x) for x in x_testdataA]))
if if_plot:
featurePlot(
energyA,
'A',
y_labelA,
title='Signal Energy of Frame (set A) vs Feature Labels (set A)')
featurePlot(
energyB,
'B',
y_labelB,
title='Signal Energy of Frame (set B) vs Feature Labels (set B)')
return energyA, energyB, ener_testA
def zero_crossing(x_dataA,
x_dataB,
x_testdataA,
y_labelA,
y_labelB,
if_plot=False):
"""zero crossing rate of frame"""
zero_crossingsA = normalize(
np.array([audioFeatureExtraction.stZCR(x) for x in x_dataA]))
zero_crossingsB = normalize(
np.array([audioFeatureExtraction.stZCR(x) for x in x_dataB]))
zc_testA = normalize(
np.array([audioFeatureExtraction.stZCR(x) for x in x_testdataA]))
if if_plot:
featurePlot(
zero_crossingsA,
'A',
y_labelA,
title='Zero Crossings Rate of Frame (set A) vs Feature Labels'
' (set A)')
featurePlot(
zero_crossingsB,
'B',
y_labelB,
title='Zero Crossings Rate of Frame (set B) vs Feature Labels'
' (set B)')
return zero_crossingsA, zero_crossingsB, zc_testA
def spectral_centroid_frame(X_dataA,
X_dataB,
X_testdataA,
y_labelA,
y_labelB,
framerate_A,
framerate_B,
if_plot=False):
"""Get Spectral centroid of frame (given abs(FFT))"""
FsA = int(framerate_A[0] / 20) # framerate is the same for all of set A
FsB = int(framerate_B[0] / 10) # framerate is the same for all of set B
centroidA = np.array([
audioFeatureExtraction.stSpectralCentroidAndSpread(np.abs(X), FsA)
for X in X_dataA
])
centroidA[:, 0] = normalize(centroidA[:, 0])
centroidA[:, 1] = normalize(centroidA[:, 1])
centroidB = np.array([
audioFeatureExtraction.stSpectralCentroidAndSpread(np.abs(X), FsB)
for X in X_dataB
])
centroidB[:, 0] = normalize(centroidB[:, 0])
centroidB[:, 1] = normalize(centroidB[:, 1])
cent_testA = np.array([
audioFeatureExtraction.stSpectralCentroidAndSpread(np.abs(X), FsA)
for X in X_testdataA
])
cent_testA[:, 0] = normalize(cent_testA[:, 0])
cent_testA[:, 1] = normalize(cent_testA[:, 1])
if if_plot:
featurePlot(
centroidA[:, 0],
'A',
y_labelA,
title='Centroids (1st Column of Set A) vs. Feature Labels (Set A)')
featurePlot(
centroidA[:, 1],
'A',
y_labelA,
title='Centroids (2nd Column of Set A) vs. Feature Labels (Set A)')
featurePlot(
centroidB[:, 0],
'B',
y_labelB,
title='Centroids (1st Column of Set B) vs. Feature Labels (Set B)')
featurePlot(
centroidB[:, 1],
'B',
y_labelB,
title='Centroids (2nd Column of Set B) vs. Feature Labels (Set B)')
return centroidA, centroidB, cent_testA
def spectral_flux(X_dataA,
X_dataB,
X_testdataA,
y_labelA,
y_labelB,
if_plot=False):
""" Get spectral flux """
fluxA = normalize(
np.array([
np.abs(
audioFeatureExtraction.stSpectralFlux(X[:int(len(X) / 2)],
X[int(len(X) / 2):]))
for X in X_dataA
]))
fluxB = normalize(
np.array([
np.abs(
audioFeatureExtraction.stSpectralFlux(X[:int(len(X) / 2)],
X[int(len(X) / 2) + 1:]))
for X in X_dataB
]))
flux_testA = normalize(
np.array([
np.abs(
audioFeatureExtraction.stSpectralFlux(X[:int(len(X) / 2)],
X[int(len(X) / 2):]))
for X in X_testdataA
]))
if if_plot:
featurePlot(fluxA,
'A',
y_labelA,
title='Spectral Flux (set A) vs Feature Labels (set A)')
featurePlot(fluxB,
'B',
y_labelB,
title='Spectral Flux (set B) vs Feature Labels (set B)')
return fluxA, fluxB, flux_testA
def entropy_of_energy(x_dataA,
x_dataB,
x_testdataA,
y_labelA,
y_labelB,
if_plot=False):
"""Entropy of Energy"""
entropyA = normalize(
np.array([
audioFeatureExtraction.stEnergyEntropy(x, numOfShortBlocks=50)
for x in x_dataA
]))
entropyB = normalize(
np.array([
audioFeatureExtraction.stEnergyEntropy(x, numOfShortBlocks=50)
for x in x_dataB
]))
entr_testA = normalize(
np.array([
audioFeatureExtraction.stEnergyEntropy(x, numOfShortBlocks=50)
for x in x_testdataA
]))
if if_plot:
featurePlot(
entropyA,
'A',
y_labelA,
title='Entropy of Energy (set A) vs Feature Labels (set A)')
featurePlot(
entropyB,
'B',
y_labelB,
title='Entropy of Energy (set B) vs Feature Labels (set B)')
return entropyA, entropyB, entr_testA
def get_mcfcc_feat(x_dataA,
x_dataB,
x_testdataA,
framerate_A,
framerate_B,
if_plot=False):
"""Get the MFCC features"""
FsA = int(framerate_A[0] / 10) # framerate is the same for all of set A
FsB = int(framerate_B[0] / 4) # framerate is the same for all of set B
mfccA = np.array([
mfcc(np.abs(x), samplerate=FsA, numcep=3, winlen=0.025)
for x in x_dataA
])
mfccA_feat = np.array([[
np.average(m[:, 0]),
np.std(m[:, 0]),
np.average(m[:, 1]),
np.std(m[:, 1]),
np.average(m[:, 2]),
np.std(m[:, 2])
] for m in mfccA])
for i in range(len(mfccA_feat[0])):
mfccA_feat[:, i] = np.abs(mfccA_feat[:, i]) / max(
np.abs(mfccA_feat[:, i]))
mfccB = np.array([
mfcc(np.abs(x), samplerate=FsB, numcep=5, winlen=0.01, winstep=0.01)
for x in x_dataB
])
mfccB_feat = np.array([[
np.average(m[:, 0]),
np.std(m[:, 0]),
np.average(m[:, 1]),
np.std(m[:, 1]),
np.average(m[:, 2]),
np.std(m[:, 2])
] for m in mfccB])
for i in range(len(mfccB_feat[0])):
mfccB_feat[:, i] = np.abs(mfccB_feat[:, i]) / max(
np.abs(mfccB_feat[:, i]))
mfcc_testA = np.array([
mfcc(np.abs(x), samplerate=FsA, numcep=3, winlen=0.025)
for x in x_testdataA
])
mfcctestA_feat = np.array([[
np.average(m[:, 0]),
np.std(m[:, 0]),
np.average(m[:, 1]),
np.std(m[:, 1]),
np.average(m[:, 2]),
np.std(m[:, 2])
] for m in mfcc_testA])
for i in range(len(mfccA_feat[0])):
mfccA_feat[:, i] = np.abs(mfccA_feat[:, i]) / max(
np.abs(mfccA_feat[:, i]))
if if_plot:
featurePlot(
mfccA_feat[:, 0],
'A',
title=
'MFCC Features (1st column) (Set A) vs Features Labels (Set A)')
featurePlot(
mfccA_feat[:, 1],
'A',
title=
'MFCC Features (2nd column) (Set A) vs Features Labels (Set A)')
featurePlot(
mfccB_feat[:, 0],
'B',
title=
'MFCC Features (1st column) (Set B) vs Features Labels (Set B)')
featurePlot(
mfccB_feat[:, 1],
'B',
title=
'MFCC Features (2nd column) (Set B) vs Features Labels (Set B)')
return mfccA_feat, mfccB_feat, mfcctestA_feat
def set_A(x_dataA, x_testdataA, y_labelA, if_plot=False):
"""Get peaks for A"""
dataA_peaks = find_peaks(x_dataA, 'A')
testdataA_peaks = find_peaks(x_testdataA, 'A')
# Get the S1 and S2 bounds for set A
s1_boundsA, s2_boundsA = get_S1S2_bounds(x_dataA, dataA_peaks, 'A')
s1_boundstA, s2_boundstA = get_S1S2_bounds(x_testdataA, testdataA_peaks,
'A')
# Standard deviation of S1
stdS1_A = normalize(stdInterval(s1_boundsA, 0, s1_boundsA, 1, x_dataA))
stdS1_testA = normalize(
stdInterval(s1_boundstA, 0, s1_boundstA, 1, x_testdataA))
# Standard deviation of S2
stdS2_A = normalize(stdInterval(s2_boundsA, 0, s2_boundsA, 1, x_dataA))
stdS2_testA = normalize(
stdInterval(s2_boundstA, 0, s2_boundstA, 1, x_testdataA))
# frequency intervals of S1 and S2
freqS1_A = freqInterval(x_dataA, s1_boundsA, 0, s1_boundsA, 1)
freqS2_A = freqInterval(x_dataA, s2_boundsA, 0, s2_boundsA, 1)
# standard deviation of S1 and S2 frequencies
stdS1_freqA = normalize(np.array([np.std(f) for f in freqS1_A]))
stdS2_freqA = normalize(np.array([np.std(f) for f in freqS2_A]))
# mean of S1 and S2 frequencies
meanS1_freqA = normalize(np.array([np.average(f) for f in freqS1_A]))
meanS2_freqA = normalize(np.array([np.average(f) for f in freqS2_A]))
if if_plot:
featurePlot(
stdS1_A,
'A',
y_labelA,
title='Standard Deviation of S1 (set A) vs. Feature Labels (set A)'
)
featurePlot(
stdS2_A,
'A',
y_labelA,
title='Standard Deviation of S2 (set A) vs. Feature Labels (set A)'
)
featurePlot(
stdS1_freqA,
'A',
y_labelA,
title=
'Standard Deviation of Frequency of S1 (set A) vs. Feature Labels'
' (set A)')
featurePlot(
meanS1_freqA,
'A',
y_labelA,
title='Mean of Frequency of S1 (set A) vs. Feature Labels (set A)')
featurePlot(
stdS2_freqA,
'A',
y_labelA,
title=
'Standard Deviation of Frequency of S2 (set A) vs. Feature Labels'
' (set A)')
featurePlot(
meanS2_freqA,
'A',
y_labelA,
title='Mean of Frequency of S2 (set A) vs. Feature Labels (set A)')
n = 2
plt.figure()
plt.title(y_labelA[n])
plt.plot(x_dataA[n], 'b')
plt.scatter(dataA_peaks[n], x_dataA[n][dataA_peaks[n]], c='r')
# plt.show()
plt.savefig(y_labelA[n] + '.png')
return stdS1_A, stdS1_testA, stdS2_A, stdS2_testA, meanS1_freqA, meanS2_freqA, stdS1_freqA, stdS2_freqA
def set_B(x_dataB, y_labelB, if_plot=False):
"""get peaks for B"""
dataB_peaks = find_peaks(x_dataB, 'B')
# Get S1 and S2 bounds for set B
s1_boundsB, s2_boundsB = get_S1S2_bounds(x_dataB, dataB_peaks, 'B')
# Standard deviation of S1
stdS1_B = normalize(stdInterval(s1_boundsB, 0, s1_boundsB, 1, x_dataB))
# Standard deviation of S2
stdS2_B = normalize(stdInterval(s2_boundsB, 0, s2_boundsB, 1, x_dataB))
# frequency intervals of S1 and S2
freqS1_B = freqInterval(x_dataB, s1_boundsB, 0, s1_boundsB, 1)
freqS2_B = freqInterval(x_dataB, s2_boundsB, 0, s2_boundsB, 1)
# standard deviation of S1 and S2 frequencies
stdS1_freqB = normalize(np.array([np.std(f) for f in freqS1_B]))
stdS2_freqB = normalize(np.array([np.std(f) for f in freqS2_B]))
# mean of S1 and S2 frequencies
meanS1_freqB = normalize(np.array([np.average(f) for f in freqS1_B]))
meanS2_freqB = normalize(np.array([np.average(f) for f in freqS2_B]))
if if_plot:
featurePlot(
stdS1_B,
'B',
y_labelB,
title='Standard Deviation of S1 (set B) vs. Feature Labels (set B)'
)
featurePlot(
stdS2_B,
'B',
y_labelB,
title='Standard Deviation of S2 (set B) vs. Feature Labels (set B)'
)
featurePlot(
stdS1_freqB,
'B',
y_labelB,
title=
'Standard Deviation of Frequency of S1 (set B) vs. Feature Labels'
' (set B)')
featurePlot(
stdS2_freqB,
'B',
y_labelB,
title=
'Standard Deviation of Frequency of S2 (set B) vs. Feature Labels'
' (set B)')
featurePlot(
meanS1_freqB,
'B',
y_labelB,
title='Mean of Frequency of S1 (set B) vs. Feature Labels (set B)')
featurePlot(
meanS2_freqB,
'B',
y_labelB,
title='Mean of Frequency of S2 (set B) vs. Feature Labels (set B)')
n = 390
plt.figure()
plt.title(y_labelB[n])
plt.plot(x_dataB[n], 'b')
plt.scatter(dataB_peaks[n], x_dataB[n][dataB_peaks[n]], c='r')
plt.show()
return stdS1_B, stdS2_B, meanS1_freqB, meanS2_freqB, stdS1_freqB, stdS2_freqB