import numpy as np
import sys
sys.path.append('../')
from datagrabber import extractAndSave, extractAndSaveYoutubeData
IEMOCAP_LOCATION = "../../../../local"
YOUTUBE_LOCATION = "../../../../local/wild_dataset/10_to_20_seconds"
def amplitude(frame, filename):
'''
Returns the maxmimum, average and variance amplitude of a frame
'''
return [np.amax(frame), np.average(frame), np.var(frame)]
# Extract amplitude from IEMOCAP and YouTube datasets
extractAndSave(amplitude, ["max", "mean", "var"], IEMOCAP_LOCATION, 2, True,
True)
extractAndSaveYoutubeData(amplitude, ["max", "mean", "var"], YOUTUBE_LOCATION,
2, True, True)
greatest_amp = (spectrum.peaks()[0])[0]
#a counter will be used to iterate over the positions of the list
list_pos = 0
#the pairs with amplitudes which are within the threshold will be copied to a new list
selected_pairs = []
while True:
#select the next pair from the list
pair = (spectrum.peaks()[list_pos])
#test whether the pair's amplitude is within the threshold
within_threshold = (pair[0] > (greatest_amp * threshold))
#if not within the threshold the loop can break as all followings pairs will have lower amplitudes
if within_threshold == False:
break
#if within the threshold add the pair to the list
selected_pairs.append(pair)
#increment the counter for the loop
list_pos = list_pos + 1
# selected_pairs is now a list of the pairs which have a prominent amplitude
sorted_by_Hz = sorted(selected_pairs, key=lambda tup: tup[1])
# the lowest Hz should now be F0
if len(sorted_by_Hz) == 0:
return [0]
else:
return [sorted_by_Hz[0][1]]
# Extract f0 from IEMOCAP and YouTube datasets
extractAndSave(f0, ['f0'], IEMOCAP_LOCATION, 2, True, True)
extractAndSaveYoutubeData(f0, ["f0"], YOUTUBE_LOCATION, 2, True, True)
Created on Fri Nov 11 12:18:26 2016
@author: Olly Styles
"""
import sys
sys.path.append('../')
from datagrabber import extractAndSave
from datagrabber import extractAndSave, extractAndSaveYoutubeData
IEMOCAP_LOCATION = "../../../../local"
YOUTUBE_LOCATION = "../../../../local/wild_dataset/10_to_20_seconds"
def zerocrossing(frame, audiofile):
'''
Returns the number of times the signal crosses the y-axis for a given
frame.
'''
n = 0
for i in range(0, len(frame) - 1):
if (frame[i] > 0 != frame[i + 1] > 0): # != is xor operator in python
n += 1
return [n]
# Extract energy from IEMOCAP and YouTube datasets
extractAndSave(zerocrossing, ['zerocrossing'], IEMOCAP_LOCATION, 2, True, True)
extractAndSaveYoutubeData(zerocrossing, ["zerocrossing"], YOUTUBE_LOCATION, 2,
True, True)
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Created on Sat Nov 12 12:04:22 2016
@author: Tom
"""
import sys
sys.path.append('../')
from datagrabber import extractAndSave, extractAndSaveYoutubeData
import numpy as np
IEMOCAP_LOCATION = "../../../../local"
YOUTUBE_LOCATION = "../../../../local/wild_dataset/10_to_20_seconds"
def energy(frame, audiofile):
'''
Return energy of a frame, defined as the sum of squared amplitudes
'''
return [sum(np.apply_along_axis(lambda x: x**2, 0, frame))]
# Extract energy from IEMOCAP and YouTube datasets
extractAndSave(energy, ['energy'], IEMOCAP_LOCATION, 2, True, True)
extractAndSaveYoutubeData(energy, ["energy"], YOUTUBE_LOCATION, 2, True, True)
m = aub.mfcc(fftsize, 40, coefficientsCount, sampleRate)
#first we need to convert this frame to the power spectrum using a DFT
p = aub.pvoc(fftsize, int(frame_size))
#in order to compute DFT the frame must be of a length which is a power of 2, so expand to fftsize using zero padding
if len(frame) != 16000:
frame = np.pad(frame, (0, frame_size - len(frame)),
'constant',
constant_values=0)
#compute the power spectrum
spec = p(frame.astype(np.float32))
#compute the MFCC, which returns the coefficents of each of the 12 coefficents
mfcc_out = m(spec)
return mfcc_out
# 1. Frame the signal into short frames.
# 2. For each frame calculate the periodogram estimate of the power spectrum.
# 3. Apply the mel filterbank to the power spectra, sum the energy in each filter.
# 4. Take the logarithm of all filterbank energies.
# 5. Take the DCT of the log filterbank energies.
# 6. Keep DCT coefficients 2-13, discard the rest
#http://practicalcryptography.com/miscellaneous/machine-learning/guide-mel-frequency-cepstral-coefficients-mfccs/#computing-the-mel-filterbank
# Extract MFCC from IEMOCAP and YouTube datasets
extractAndSave(mfcc, labels, IEMOCAP_LOCATION, 2, True, True)
extractAndSaveYoutubeData(mfcc, labels, YOUTUBE_LOCATION, 2, True, True)
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Created on Sat Nov 12 12:04:22 2016
@author: Tom
"""
import sys
sys.path.append('../sourceFiles/')
import thinkplot as tp
import thinkdsp as td
sys.path.append('../')
from datagrabber import extractAndSave, extractAndSaveYoutubeData
IEMOCAP_LOCATION = "../../../../local"
YOUTUBE_LOCATION = "../../../../local/wild_dataset/10_to_20_seconds"
#make_spectrum is sorted in order of decreasing amplitude, so the first pair is going to be the amplitude of the pitch and the pitch frequency.
def pitch(frame, audiofile):
clip = td.Wave(frame)
spectrum = clip.make_spectrum()
return [spectrum.peaks()[0][1]]
#extractAndSave(pitch,['pitch'],IEMOCAP_LOCATION,2,False)
extractAndSaveYoutubeData(pitch, ["pitch"], YOUTUBE_LOCATION, 2)
sys.path.append('../')
from datagrabber import extractAndSave,extractAndSaveYoutubeData
IEMOCAP_LOCATION = "../../../../local"
YOUTUBE_LOCATION = "../../../../local/wild_dataset/10_to_20_seconds"
def cepstrum(frame, filename):
'''
Computes the cepstrum of the frame, which is calculated by converting
the power spectrum (|DCT|^2) of the audio frame to a log-scale.
The max average and variance of the resulting audio clip are returned
as features
'''
#get DCT
audio = np.fft.fft(frame)
#get |DCT|
audio = abs(audio)
#get |DCT|^2, i.e. the power spectrum
audio = audio ** 2
#take a log, which simulates the non-linearality of human hearing
audio = np.log2(audio)
#take the real of the IDFT as the feature
audio = np.fft.ifft(audio)
audio = audio.real
return [np.amax(audio), np.average(audio), np.var(audio)]
# Extract cepstrum from IEMOCAP and YouTube datasets
extractAndSave(cepstrum,["max", "mean", "var"],IEMOCAP_LOCATION,2,True,True)
extractAndSaveYoutubeData(cepstrum,["max", "mean","var"],YOUTUBE_LOCATION,2,True,True)