def logMelSpectrum(input, samplingrate):
"""
Calculates the log output of a Mel filterbank when the input is the power spectrum
Args:
input: array of power spectrum coefficients [N x nfft] where N is the number of frames and
nfft the length of each spectrum
samplingrate: sampling rate of the original signal (used to calculate the filterbank shapes)
Output:
array of Mel filterbank log outputs [N x nmelfilters] where nmelfilters is the number
of filters in the filterbank
Note: use the trfbank function provided in lab1_tools.py to calculate the filterbank shapes and
nmelfilters
"""
filter_bank = trfbank(samplingrate, input.shape[1])
#uncomment to display filter bank plots
# _ = plt.rcParams['figure.figsize'] = [15, 3]
# _ = plt.subplot(211)
# _ = plt.plot(filter_bank)
# _ = plt.subplot(212)
# _ = plt.plot(filter_bank.T)
# plt.show()
#too much variation from the example
#without doing this there were only handful of variations from example
#filter_bank = np.where(filter_bank == 0, np.finfo(float).eps, filter_bank) # Numerical Stability
return np.log(input.dot(filter_bank.T))
def logMelSpectrum(input, samplingrate):
"""
Calculates the log output of a Mel filterbank when the input is the power spectrum
Args:
input: array of power spectrum coefficients [N x nfft] where N is the number of frames and
nfft the length of each spectrum
samplingrate: sampling rate of the original signal (used to calculate the filterbank shapes)
Output:
array of Mel filterbank log outputs [N x nmelfilters] where nmelfilters is the number
of filters in the filterbank
Note: use the trfbank function provided in lab1_proto.py to calculate the filterbank shapes and
nmelfilters
"""
N, nfft = input.shape # 92frames
Mel = trfbank(samplingrate, nfft) # 40filters*512
M = Mel.shape[0]
logMelSpec = np.zeros((N, M))
for i in range(N):
for j in range(M):
logMelSpec[i, j] = np.log(np.sum(input[i] * Mel[j]))
return logMelSpec
def logMelSpectrum(input, samplingrate):
"""
Calculates the log output of a Mel filterbank when the input is the power spectrum
Args:
input: array of power spectrum coefficients [N x nfft] where N is the number of frames and
nfft the length of each spectrum
samplingrate: sampling rate of the original signal (used to calculate the filterbank shapes)
Output:
array of Mel filterbank log outputs [N x nmelfilters] where nmelfilters is the number
of filters in the filterbank
Note: use the trfbank function provided in lab1_tools.py to calculate the filterbank shapes and
nmelfilters
"""
fil_bank = trfbank(samplingrate,
512,
lowfreq=133.33,
linsc=200 / 3.,
logsc=1.0711703,
nlinfilt=13,
nlogfilt=27,
equalareas=False)
mspec = np.log(np.dot(input, np.transpose(fil_bank)))
#plt.pcolormesh(mspecc)
return mspec
def logMelSpectrum(input, samplingrate):
"""
Calculates the log output of a Mel filterbank when the input is the power spectrum
Args:
input: array of power spectrum coefficients [N x nfft] where N is the number of frames and
nfft the length of each spectrum
samplingrate: sampling rate of the original signal (used to calculate the filterbank shapes)
Output:
array of Mel filterbank log outputs [N x nmelfilters] where nmelfilters is the number
of filters in the filterbank
Note: use the trfbank function provided in lab1_tools.py to calculate the filterbank shapes and
nmelfilters
"""
return np.log(input.dot(trfbank(samplingrate, input.shape[1]).T))
def logMelSpectrum(input, samplingrate, nfft):
"""
Calculates the log output of a Mel filterbank when the input is the power spectrum
Args:
input: array of power spectrum coefficients [N x nfft] where N is the number of frames and
nfft the length of each spectrum
samplingrate: sampling rate of the original signal (used to calculate the filterbank shapes)
Output:
array of Mel filterbank log outputs [N x nmelfilters] where nmelfilters is the number
of filters in the filterbank
Note: use the trfbank function provided in lab1_tools.py to calculate the filterbank shapes and
nmelfilters
"""
import matplotlib.pyplot as plt
t = trfbank(samplingrate, nfft)
if (False): # add flag
plt.plot(t)
plt.title('Trfbank')
plt.show()
return np.log(np.dot(input, t.T))
def logMelSpectrum(input, samplingrate):
"""
Calculates the log output of a Mel filterbank when the input is the power spectrum
Args:
input: array of power spectrum coefficients [N x nfft] where N is the number of frames and
nfft the length of each spectrum
samplingrate: sampling rate of the original signal (used to calculate the filterbank shapes)
Output:
array of Mel filterbank log outputs [N x nmelfilters] where nmelfilters is the number
of filters in the filterbank
Note: use the trfbank function provided in lab1_tools.py to calculate the filterbank shapes and
nmelfilters
"""
N = len(input) # 92frames
Mel = lab1_tools.trfbank(samplingrate, len(input[0])) # 40filters*512
M = Mel.shape[0]
# plot the filters in linear frequency scale x=k*512/fs(20000)
#plt.figure()
x = np.zeros((N, M))
for i in range(N):
for j in range(M):
x[i, j] = np.log(np.sum(input[i, :] * Mel[j, :]))
return x
freq = np.linspace(-0.5, 0.5, len(A))
response = np.abs(fftshift(A / abs(A).max()))
plt.plot(freq, response)
plt.title("Frequency response of the Hamming window")
plt.ylabel("Frequency space amplitude")
plt.xlabel("Frequency space sample")
plt.savefig("./plots/hamming_fft.png", bbox_inches='tight')
# ===================================================================
fig, ax = plt.subplots(figsize=fig_size)
ax.set_title("Power spectrum")
ax.pcolormesh(example['spec'].T)
fig.savefig("./plots/power_spectrum.png", bbox_inches='tight')
# ===========================================================
# 4.5
filter_ = trfbank(example['samplingrate'], 512)
plt.figure(figsize=(12, 2))
plt.xlim(0, 180)
for cap in filter_:
plt.plot(cap)
plt.title("Mel filterbank")
plt.savefig("./plots/mel_filterbank.png", bbox_inches='tight')
#
fig, ax = plt.subplots(figsize=fig_size)
ax.set_title("log-scale Mel spectrum")
ax.pcolormesh(example['mspec'].T)
fig.savefig("./plots/log_mel_spectrum.png", bbox_inches='tight')
# ====================================================================
# 4.6
fig, ax = plt.subplots(figsize=fig_size)
