def extract(self, file_name):
fs, signal = wav.read(file_name)
# print(file_name)
signal = signal[:, 0]
mfcc = speechpy.mfcc(signal,
sampling_frequency=fs,
frame_length=0.020,
frame_stride=0.01,
num_filters=40,
fft_length=512,
low_frequency=0,
high_frequency=None)
#print(mfcc.shape)
#print(mfcc)
mfcc_feature_cube = speechpy.extract_derivative_feature(mfcc)
#print('mfcc feature cube shape:' , mfcc_feature_cube.shape)
#print(mfcc_feature_cube)
return mfcc, mfcc_feature_cube
lib_path = os.path.abspath(os.path.join('..'))
print(lib_path)
sys.path.append(lib_path)
import speechpy
import os
file_name = os.path.join(os.path.dirname(os.path.abspath(__file__)),
'Alesis-Sanctuary-QCard-AcoustcBas-C2.wav')
fs, signal = wav.read(file_name)
signal = signal[:, 0]
############# Extract MFCC features #############
mfcc = speechpy.mfcc(signal,
sampling_frequency=fs,
frame_length=0.020,
frame_stride=0.01,
num_filters=40,
fft_length=512,
low_frequency=0,
high_frequency=None)
mfcc_cmvn = speechpy.cmvnw(mfcc, win_size=301, variance_normalization=True)
print('mfcc(mean + variance normalized) feature shape=', mfcc_cmvn.shape)
mfcc_feature_cube = speechpy.extract_derivative_feature(mfcc)
print('mfcc feature cube shape=', mfcc_feature_cube.shape)
############# Extract logenergy features #############
logenergy = speechpy.lmfe(signal,
sampling_frequency=fs,
frame_length=0.020,
frame_stride=0.01,
num_filters=40,
import time
from speechpy import mfcc
from speechpy import delta
from speechpy import log_filter_bank
import scipy.io.wavfile as wav
from scipy import signal as sig
import matplotlib.pyplot as plt
from random import shuffle
if __name__ == '__main__':
filename = './sample_24414.wav'
rate, signal = wav.read(filename)
mfcc_feature = mfcc(signal, rate)
d_mfcc_feature = delta(mfcc_feature, 2)
filter_bank_feature = log_filter_bank(signal, rate)
feature = filter_bank_feature[1:3, :]
print('signal:', signal)
print('rate:', rate)
print('np.shape(signal):', np.shape(signal))
print('mfcc_feature:', mfcc_feature)
print('np.shape(mfcc_feature):', np.shape(mfcc_feature))
print('d_mfcc_feature:', d_mfcc_feature)
print('np.shape(d_mfcc_feature):', np.shape(d_mfcc_feature))
print('feature:', feature)
print('np.shape(feature):', np.shape(feature))
# exit(0)
def get_mfcc(filename,
downsample=0,
delta=False,
noisereduction=True,
normalizemean=False,
numcoeff=13,
verbose=False):
def print_if(string, verb):
if verb:
print(string)
""" Returns the MFCC of a given WAV file as a numpy array. Options include:
delta: append delta (velocity) features to MFCC; doubles # features per frame
noisereduction: apply noise reduction before computing MFCC
normalizemean: output MFCC as normalized global mean - mutually exclusive from delta
numcoeff: specify number of cepstral coefficients; usually scaled linearly with sampling rate
verbose: enable detailed print statements
"""
# Perform downsampling and creates another downsampled wav file if specified
if downsample > 0:
print_if('Downsampling On by factor of ' + str(downsample), verbose)
resample(filename, downsample)
filename = "resampled.wav"
else:
print_if('Downsampling Off', verbose)
# Perform noise reduction before calculating coefficients
if noisereduction:
print_if('Noise Reduction On', verbose)
fs, signal = reduce_noise(filename)
else:
print_if('Noise Reduction Off', verbose)
fs, signal = wav.read(filename)
print_if('\nMFCC (Mel Frequency Cepstral Coefficients)\n' + div, verbose)
print_if('File sampling frequency: ' + str(fs) + '\n', verbose)
mfcc = speechpy.mfcc(
signal,
sampling_frequency=fs,
frame_length=0.020,
frame_stride=0.020, # Frame overlap amount (0.02 is no overlap)
num_cepstral=numcoeff, # Default 13; scale with sample rate
num_filters=40,
fft_length=512,
low_frequency=0,
high_frequency=None,
dc_elimination=True)
mfcc_feature_cube = speechpy.extract_derivative_feature(mfcc)
length, numfeatures, dim = mfcc_feature_cube.shape
print_if('MFCC dimension: ' + str(mfcc_feature_cube.shape), verbose)
print_if(mfcc, verbose)
if delta:
delta_features = get_delta(mfcc)
mfcc_delta = np.empty((length, numfeatures * 2))
for i in range(len(mfcc)):
mfcc_delta[i] = np.concatenate([mfcc[i], delta_features[i]])
mfcc_delta_feature_cube = speechpy.extract_derivative_feature(
mfcc_delta)
print_if(
'\nMFCC with Delta dimension: ' +
str(mfcc_delta_feature_cube.shape), verbose)
print_if(mfcc_delta, verbose)
return mfcc_delta
# Option to return normalized cepstral mean (for each set of coeffs, subtracts mean from each coeff)
if normalizemean:
mfcc_normalizedmean = speechpy.cmvn(mfcc, variance_normalization=False)
mfcc_normalizedmean_feature_cube = speechpy.extract_derivative_feature(
mfcc_normalizedmean)
print_if(
'\nMFCC with normalized mean dimension: ' +
str(mfcc_normalizedmean_feature_cube.shape), verbose)
print_if(mfcc_normalizedmean, verbose)
return mfcc_normalizedmean
return mfcc