def hammingWindow():
'''
Generate plot of a Hamming window and windowed signal.
'''
d = np.linspace(0, 4 * np.pi, 200)
h = 200 * hamming(200)
sig = 100 * np.sin(3 * d)
windowed = sig * h / 200
plt.plot(h)
plt.xticks([])
plt.yticks([])
plt.savefig('HammingWindowFunction.pdf')
plt.close()
plt.plot(sig)
plt.xticks([])
plt.yticks([])
plt.savefig('Original.pdf')
plt.close()
plt.plot(windowed)
plt.xticks([])
plt.yticks([])
plt.savefig('WindowedSignal.pdf')
plt.close()
def hammingWindow():
'''
Generate plot of a Hamming window and windowed signal.
'''
d = np.linspace(0,4*np.pi, 200)
h = 200*hamming(200)
sig = 100*np.sin(3*d)
windowed = sig*h/200
plt.plot(h)
plt.xticks([])
plt.yticks([])
plt.savefig('HammingWindowFunction.pdf')
plt.close()
plt.plot(sig)
plt.xticks([])
plt.yticks([])
plt.savefig('Original.pdf')
plt.close()
plt.plot(windowed)
plt.xticks([])
plt.yticks([])
plt.savefig('WindowedSignal.pdf')
plt.close()
def extract(x):
'''
Extract MFCC coefficients of the sound signal x in numpy array format.
Inputs:
x -- array of shape (88200,)
Returns:
array of shape (198,10), the MFCCs of x.
'''
window = hamming(1323)
feature = []
for i in xrange(198):
# Windowing
frame = x[i * 441:i * 441 + 1323] * window
# Pre-emphasis
frame[1:] -= frame[:-1] * .95
# Power spectrum
X = np.abs(pyfftw.interfaces.scipy_fftpack.fft(frame,
n=2048)[:1025])**2
X[X < 1e-100] = 1e-100 # Avoid zero
# Mel filtering, logarithm, DCT
M = melfb(40, 2048, 44100)
X = 0.25 * pyfftw.interfaces.scipy_fftpack.dct(np.log(np.dot(M,
X)))[1:11]
feature.append(X)
feature = np.row_stack(feature)
return feature
def window(f):
'''
Break up a signal into frames, and multiply each frame with a Hamming window.
Create 198 frames of length 1323, with adjacent frames overlapping by 882 values.
Inputs:
f -- array of shape (88200,)
Returns:
List of length 198, containing arrays of shape (1323,), ordered
according to the position of the frames relative to f.
'''
window = hamming(1323)
frames = []
for i in xrange(198):
frame = f[i * 441:i * 441 + 1323] * window
frames.append(frame)
return frames
def window(f):
'''
Break up a signal into frames, and multiply each frame with a Hamming window.
Create 198 frames of length 1323, with adjacent frames overlapping by 882 values.
Inputs:
f -- array of shape (88200,)
Returns:
List of length 198, containing arrays of shape (1323,), ordered
according to the position of the frames relative to f.
'''
window = hamming(1323)
frames = []
for i in xrange(198):
frame = f[i*441: i*441 + 1323]*window
frames.append(frame)
return frames
def extract(x):
'''
Extract MFCC coefficients of the sound signal x in numpy array format.
Inputs:
x -- array of shape (88200,)
Returns:
array of shape (198,10), the MFCCs of x.
'''
window = hamming(1323)
feature = []
for i in xrange(198):
# Windowing
frame = x[i*441: i*441 + 1323]*window
# Pre-emphasis
frame[1:] -= frame[:-1] * .95
# Power spectrum
X = np.abs(pyfftw.interfaces.scipy_fftpack.fft(frame, n=2048)[:1025]) ** 2
X[X < 1e-100] = 1e-100 # Avoid zero
# Mel filtering, logarithm, DCT
M = melfb(40, 2048, 44100)
X = 0.25*pyfftw.interfaces.scipy_fftpack.dct(np.log(np.dot(M,X)))[1:11]
feature.append(X)
feature = np.row_stack(feature)
return feature
