69. Preprocessing

The dataset includes audio files (wav, mp3, etc) of male and female orgasm.

This notebook is for preprocessing, feature engineering etc.

import numpy as np
import matplotlib.pyplot as plt
import matplotlib.colors as colors
plt.style.use('seaborn-poster')
    
import scipy as sp
from scipy import signal
from scipy.io.wavfile import read
import sys
import os

import soundfile as sf

import librosa
import librosa.display

path = '../../../Downloads/dataset/'
files = os.listdir(path)

wavfiles = []

for file in files:
    if file[-3:]=='wav':
        wavfiles.append(file)
    

wavfiles[0]

'FEET LOVE.wav'

path+wavfiles[0]

'../../../Downloads/dataset/FEET LOVE.wav'

testfile=path+wavfiles[2]
def dataext(file):
    
    data = sf.read(file)
    fs = data[1]
    audio = data[0]#[:50000]
    audio = np.transpose(audio)*100
    # audio[np.where(audio < 1)] = 1
    # audio = 10 * np.log10(audio)
    if len(audio) == 2:
        amp = [np.max(audio[0]), np.max(audio[1]) ]
        time = np.arange( len(audio[0]) ) / float(fs)
    elif len(audio) == 1 or len(audio) > 100:
        amp = np.max(audio)
        time = np.arange( len(audio) ) / float(fs)
    elif len(audio) == 3:
        amp = [np.max(audio[0]), np.max(audio[1]), np.max(audio[2]) ]
        time = np.arange( len(audio[0]) ) / float(fs)
    elif len(audio) == 4:
        amp = [np.max(audio[0]), np.max(audio[1]), np.max(audio[2]), np.max(audio[2]) ]
        time = np.arange( len(audio[0]) ) / float(fs)

        
    return [amp, time, fs, audio ]

dataext(testfile)

[[80.0048828125, 80.0048828125],
 array([0.00000000e+00, 2.26757370e-05, 4.53514739e-05, ...,
        4.49537642e+01, 4.49537868e+01, 4.49538095e+01]),
 44100,
 array([[ 0.        ,  0.        ,  0.        , ..., -0.00915527,
         -0.00915527, -0.00915527],
        [ 0.        ,  0.        ,  0.        , ..., -0.00915527,
         -0.00915527, -0.00915527]])]

70. Feature Extraction

ZCR

testfiledata = dataext(testfile)

zcr = librosa.feature.zero_crossing_rate( testfiledata[-1][0] )

plt.plot( zcr[0] )
plt.show()

Mel scale spectrogram

# D = np.abs(librosa.stft( testfiledata[-1][0]))**2
# S = librosa.feature.melspectrogram(S=D)

S = librosa.feature.melspectrogram(y=testfiledata[-1][0], sr=testfiledata[-2], n_mels=128, fmax=20000)

plt.figure( figsize=(10,4) )

librosa.display.specshow( librosa.power_to_db(S, ref=np.max), y_axis='mel', fmax= 20000, x_axis='time')
plt.colorbar(format='%+2.0f dB')
plt.tight_layout()

71. Visualization

Read in data

wavaudioset = [ dataext(path+wf) for wf in wavfiles]

Define a visualization function

def vis(audiodata):
    
    amp = audiodata[0]
    time = audiodata[1]
    fs = audiodata[2]
    audio = audiodata[3]
    
    f, axarr = plt.subplots(2, sharex=True,figsize=(15,15))
#     f.suptitle('Sharing X axis')

    if len(audio) == 2:
        axarr[0].plot(time, audio[0],'r-',label="left")
        axarr[0].plot(time, audio[1],'k--',label='right')
        axarr[0].set(ylabel="Amplitude")
    #     axarr[0].xlabel("Time (s)")
        axarr[0].set(title="Waveform of Orgasm")
        axarr[0].legend()


        # fact= 20
        # audiolen = len(audio[0])
        # winlen = audiolen/fact
        winlen=512
        # norm=colors.LogNorm(vmin=Z1.min(), vmax=Z1.max()),
        #                    cmap='PuBu_r'
        f, t, Zxx = signal.stft(audio[0], fs, nperseg=winlen)
        axarr[1].pcolormesh(t, f, np.abs(Zxx), vmin=0, vmax=amp[0], cmap='hot')
        # plt.pcolormesh(t, f, np.abs(Zxx), vmin=0, vmax=ampl, norm=colors.LogNorm(vmin=0.01,vmax=10), cmap='hot')
        axarr[1].set(title='Orgasm STFT Magnitude')
        axarr[1].set(ylabel='Frequency [Hz]')
        axarr[1].set(xlabel='Time [sec]')
        axarr[1].set(ylim=[0,2000])
        # plt.xlim([0,2.5])
        
    #########
    ##!!!!!!! Have to write the code to plot monochannel data

## Test the functions
vis(wavaudioset[0])

def vismel(audiodata):
    
    amp = audiodata[0]
    time = audiodata[1]
    fs = audiodata[2]
    audio = audiodata[3]


    if len(audio) == 2:


        winlen=512

        S = librosa.feature.melspectrogram(y=audio[0], sr=fs, n_mels=128, fmax=20000)

        plt.figure( figsize=(15,4) )

        librosa.display.specshow( librosa.power_to_db(S, ref=np.max), y_axis='mel', fmax= 20000, x_axis='time')
        plt.colorbar(format='%+2.0f dB')
        plt.title('Orgasm MELSPECTROGRAM')
        plt.tight_layout()
        
    #########
    ##!!!!!!! Have to write the code to plot monochannel data

vismel(wavaudioset[0])

for item in wavaudioset:
    vismel(item)
    plt.show()

Waveform and stft

for item in wavaudioset:
    vis(item)
    plt.show()