ReadData.py

# -*- coding: utf-8 -*-
"""
Created on Tue Oct  8 16:58:36 2019

@author: Mehak
"""

#%%
import csv
from scipy.io import wavfile 
import matplotlib.pyplot as plt
import numpy as np
import os
import scipy.stats as scst
import scipy.signal as sig
import librosa

#%%

with open('Data.csv', 'r') as f:
    reader = csv.reader(f)
    all_files = list(reader)

#%%

#def analyze(data,beat_start,beat_end,fs):
#    
#    beat=data[beat_start:beat_end]
#    Pxx, freqs, times, im = plt.specgram(beat, NFFT=1024, Fs=fs,noverlap=256)
#    plt.show()
#    
#    Pxx=10*np.log10(Pxx)
#    freqmax=np.zeros((len(freqs),len(times)))
#    freqmin=np.zeros((len(freqs),len(times)))
#
#    
#    max_array=np.zeros(len(times))
#    min_array=np.multiply(np.ones(len(times)),22050)
#    range_array=np.zeros(len(times))
#    mean_array=np.zeros(len(times))
#    
#    temp=np.zeros(len(freqs))
#  
#    
#    for j in range(len(times)):
#        for i in range(len(freqs)):
#            if(Pxx[i][j]>=30):
#                freqmax[i][j]=freqs[i]
#            elif(Pxx[i][j]>=0):
#                freqmin[i][j]= freqs[i]
#            
#        max_array[j]=max(freqmax[:,j])
#        temp=freqmin[:,j]
#        temp = temp[np.nonzero(temp)]
#        if(len(temp) == 0):
#            min_array[j] = 22050
#            mean_array[j]= 0
#        else:
#            min_array[j]=min(temp)
#            mean_array[j]=np.mean(temp)        
#        
#    stat_max=max(max_array)
#    stat_min=min(min_array)
#    stat_range=stat_max-stat_min
#    stat_mean=np.mean(mean_array)
#    
#    return stat_max,stat_min,stat_range,stat_mean


#%%
#
#def index_value(fs,data):
#    window_size = 0.2*fs
#    beat = []
#    time_interval = [0, 0] #Start and end of the beat in terms of seconds
#    #print(len(data))
#    #print(window_size)
#    #print ("Number of windows: ", int((len(data))/window_size))
#    ct = 0
#    avg=np.zeros(int((len(data))/window_size))
#    for i in range(0, int((len(data))/window_size)-1):    
#        frame = data[int(np.ceil(i*window_size)):int(np.ceil((i+1)*window_size))]
#        avg[i] = np.mean(np.absolute(frame)) #Computing the average of the signal in each window frame
#     #   print ("Average", avg[i])
#    #print(max(avg))
#    #print(np.mean(avg))
#    dynamic_threshold=np.mean(avg)
#    #print(dynamic_threshold)
#    for i in range(0, int((len(data))/window_size)-1):       
#        if avg[i]>=dynamic_threshold: #Manually set threshold after observation (might have to be changed)
#     #       print(i)
#            if beat==[]:
#                time_interval[0] = (i*window_size)
#            for sample in frame:
#                beat.append(sample)
#        elif beat!=[]: #Extracting the second beat
#            ct+=1
#            if ct==1:
#                beat = []
#            else:
#                time_interval[1] = (i*window_size)
#                break
#    return int(time_interval[0]),int(time_interval[1])

#%%
def ExtractFreq(data):
    Pxx, freqs, times, im = plt.specgram(data,Fs=fs)
    
        
    intervals = len(times);
    block = np.int(np.ceil(intervals/4))
    ti1 = np.sum(Pxx[:,0:block-1], axis = 1);
    ti2 = np.sum(Pxx[:,block:2*block-1], axis = 1)
    ti3 = np.sum(Pxx[:,2*block:3*block-1], axis = 1)
    ti4 = np.sum(Pxx[:,3*block:intervals-1], axis = 1)
    
    
    Freq_vals = []
    cumsum1= np.cumsum(ti1)
    cumsum1 = cumsum1/max(cumsum1)

    flag1=0
    flag2=0
    flag3=0
    for i in range(int(len(cumsum1))):
        if(cumsum1[i]>0.95 and flag3==0):
            if(abs(cumsum1[i]-0.95)<abs(cumsum1[i-1]-0.95)):
                percent95_1=freqs[i]
            else:
                percent95_1 = freqs[i-1]
            flag3=1
        if(cumsum1[i]>0.5 and flag1==0):
            if(abs(cumsum1[i]-0.5)<abs(cumsum1[i-1]-0.5)):
                percent50_1=freqs[i]
            else:
                percent50_1 = freqs[i-1]
            flag1=1
        if(cumsum1[i]>0.05 and flag2==0):
            if(abs(cumsum1[i]-0.05)<abs(cumsum1[i-1]-0.05)):
                percent5_1=freqs[i]
            else:
                percent5_1 = freqs[i-1]
            flag2=1;
    
    Freq_vals.append(percent95_1)
    Freq_vals.append(percent50_1)
    Freq_vals.append(percent5_1)
    
    cumsum2= np.cumsum(ti2)
    cumsum2 = cumsum2/max(cumsum2)
    
    flag1=0
    flag2=0
    flag3=0
    for i in range(int(len(cumsum2))):
        if(cumsum2[i]>0.95 and flag3==0):
            if(abs(cumsum2[i]-0.95)<abs(cumsum2[i-1]-0.95)):
                percent95_2=freqs[i]
            else:
                percent95_2 = freqs[i-1]
            flag3=1
        if(cumsum2[i]>0.5 and flag1==0):
            if(abs(cumsum2[i]-0.5)<abs(cumsum2[i-1]-0.5)):
                percent50_2=freqs[i]
            else:
                percent50_2 = freqs[i-1]
            flag1=1
        if(cumsum2[i]>0.05 and flag2==0):
            if(abs(cumsum2[i]-0.05)<abs(cumsum2[i-1]-0.05)):
                percent5_2=freqs[i]
            else:
                percent5_2 = freqs[i-1]
            flag2=1;
 
    Freq_vals.append(percent95_2)
    Freq_vals.append(percent50_2)
    Freq_vals.append(percent5_2) 
    
    cumsum3= np.cumsum(ti3)
    cumsum3 = cumsum3/max(cumsum3)
    
    flag1=0
    flag2=0
    flag3=0
    for i in range(int(len(cumsum3))):
        if(cumsum3[i]>0.95 and flag3==0):
            if(abs(cumsum3[i]-0.95)<abs(cumsum3[i-1]-0.95)):
                percent95_3=freqs[i]
            else:
                percent95_3 = freqs[i-1]
            flag3=1
        if(cumsum3[i]>0.5 and flag1==0):
            if(abs(cumsum3[i]-0.5)<abs(cumsum3[i-1]-0.5)):
                percent50_3=freqs[i]
            else:
                percent50_3 = freqs[i-1]
            flag1=1
        if(cumsum3[i]>0.05 and flag2==0):
            if(abs(cumsum3[i]-0.05)<abs(cumsum3[i-1]-0.05)):
                percent5_3=freqs[i]
            else:
                percent5_3 = freqs[i-1]
            flag2=1;
    Freq_vals.append(percent95_3)
    Freq_vals.append(percent50_3)
    Freq_vals.append(percent5_3)
    
    cumsum4= np.cumsum(ti4)
    cumsum4 = cumsum4/max(cumsum4)
    
    #print(cumsum4)
    #print(freqs)
    flag1=0
    flag2=0
    flag3=0
    for i in range(int(len(cumsum4))):
        if(cumsum4[i]>0.95 and flag3==0):
            if(abs(cumsum4[i]-0.95)<abs(cumsum4[i-1]-0.95)):
                percent95_4=freqs[i]
            else:
                percent95_4 = freqs[i-1]
            flag3=1
        if(cumsum4[i]>0.5 and flag1==0):
            if(abs(cumsum4[i]-0.5)<abs(cumsum4[i-1]-0.5)):
                percent50_4=freqs[i]
            else:
                percent50_4 = freqs[i-1]
            flag1=1
            #print(percent50_4)
        if(cumsum4[i]>0.05 and flag2==0):
            if(abs(cumsum4[i]-0.05)<abs(cumsum4[i-1]-0.05)):
                percent5_4=freqs[i]
            else:
                percent5_4 = freqs[i-1]
            #print(percent5_4)
            flag2=1;
    Freq_vals.append(percent95_4)
    Freq_vals.append(percent50_4)
    Freq_vals.append(percent5_4)
    
    return Freq_vals


#%%

no_of_records = len(all_files)

for i in range(0,no_of_records):
    #Extracting each record
    record = all_files[i]
    #Getting the file name from each record extracted (first column)
    file = record[0]
    file_name = file + '_beat.wav'
    file_path = 'HandheldRecorded\\' + file_name 
    #Reading corresponding file  
    fs, audio_file = librosa.load(file_path, sr=16000)
    #fs, audio_file =  wavfile.read(file_path)
    #beat = sig.resample()
    #beat_start, beat_end = index_value(fs,audio_file)
    #maxf, minf, rangef,meanf = analyze(audio_file, beat_start, beat_end, fs)
    freq_params = ExtractFreq(audio_file)
    energy = sum(np.power(np.asarray(audio_file),2))
    record.extend(freq_params)
    record.append(energy)

#%%
with open("DataWithParams_1.csv", "w", newline ='') as f:
    writer = csv.writer(f)
    writer.writerows(all_files)

#%%

with open("DataWithParams.csv","r") as f:
    csvreader = csv.reader(f)
    all_records = list(csvreader)

maxf_array = []
meanf_array = []
rangef_array = []
psv_array = []
diameter_array = []
depth_array = []
energy_array = []

for i in range(len(all_records)):
    record = all_records[i]
    maxf_array.append(record[4])
    meanf_array.append(record[6])
    rangef_array.append(record[7])
    energy_array.append(record[8])
    psv_array.append(record[1])
    diameter_array.append(record[3])
    depth_array.append(record[2])

#%%
psv_array = np.asarray([float(i) for i in psv_array], dtype = np.float32)
diameter_array = np.asarray([float(i) for i in diameter_array], dtype = np.float32)
depth_array = np.asarray([float(i) for i in depth_array], dtype = np.float32)
maxf_array = np.asarray([float(i) for i in maxf_array], dtype = np.float32)
meanf_array = np.asarray([float(i) for i in meanf_array], dtype = np.float32)
rangef_array = np.asarray([float(i) for i in rangef_array], dtype = np.float32)
volume_array = np.multiply(psv_array,diameter_array)   
energy_array = np.asarray([float(i) for i in energy_array], dtype = np.float32) 
    
#%%

#Correlation between max frequency and psv --- 1 

covariance_1 = np.cov(maxf_array,psv_array)
corr1, _ = scst.pearsonr(maxf_array,psv_array)
corrs1,v = scst.spearmanr(maxf_array,psv_array)
print(covariance_1[0][1])
print(corr1)
print(corrs1)

#%%
#Correlation between max frequency and diameter --- 2

covariance_2 = np.cov(maxf_array,diameter_array)
corr2, _ = scst.pearsonr(maxf_array,diameter_array)
corrs2,v = scst.spearmanr(maxf_array,diameter_array)
print(covariance_2[0][1])
print(corr2)
print(corrs2)
    
#%%

#Correlation between max frequency and depth --- 3

covariance_3 = np.cov(maxf_array,depth_array)
corr3, _ = scst.pearsonr(maxf_array,depth_array)
corrs3,v = scst.spearmanr(maxf_array,depth_array)
print(covariance_3[0][1])
print(corr3)
print(corrs3)

#%%

#Correlation between max frequency and volume --- 4 

covariance_4 = np.cov(maxf_array,volume_array)
corr4, _ = scst.pearsonr(maxf_array,volume_array)
corrs4,v = scst.spearmanr(maxf_array,volume_array)
print(covariance_4[0][1])
print(corr4)
print(corrs4)

#%%
#Correlation between mean of frequency and psv --- 5 

covariance_5 = np.cov(meanf_array,psv_array)
corr5, _ = scst.pearsonr(meanf_array,psv_array)
corrs5,v = scst.spearmanr(meanf_array,psv_array)
print(covariance_5[0][1])
print(corr5)
print(corrs5)

#%%

#Correlation between mean of frequency and diameter --- 6

covariance_6 = np.cov(meanf_array,diameter_array)
corr6, _ = scst.pearsonr(meanf_array,diameter_array)
corrs6,v = scst.spearmanr(meanf_array,diameter_array)
print(covariance_6[0][1])
print(corr6)
print(corrs6)

#%%

#Correlation between mean of frequency and depth --- 7

covariance_7 = np.cov(meanf_array,depth_array)
corr7, _ = scst.pearsonr(meanf_array,depth_array)
corrs7,v = scst.spearmanr(meanf_array,depth_array)
print(covariance_7[0][1])
print(corr7)
print(corrs7)

#%%

#Correlation between mean of frequency and volume

covariance_8 = np.cov(meanf_array,volume_array)
corr8, _ = scst.pearsonr(meanf_array,volume_array)
corrs8,v = scst.spearmanr(meanf_array,volume_array)
print(covariance_8[0][1])
print(corr8)
print(corrs8)

#%%

covariance_9 = np.cov(energy_array,psv_array)
corr9, _ = scst.pearsonr(energy_array,psv_array)
corrs9,v = scst.spearmanr(energy_array,psv_array)
print(covariance_9[0][1])
print(corr9)
print(corrs9)

#%%

covariance_10 = np.cov(energy_array,diameter_array)
corr10, _ = scst.pearsonr(energy_array,diameter_array)
corrs10,v = scst.spearmanr(energy_array,diameter_array)
print(covariance_10[0][1])
print(corr10)
print(corrs10)

#%%

covariance_11 = np.cov(energy_array,depth_array)
corr11, _ = scst.pearsonr(energy_array,depth_array)
corrs11,v = scst.spearmanr(energy_array,depth_array)
print(covariance_11[0][1])
print(corr11)
print(corrs11)

#%%

covariance_12 = np.cov(energy_array,volume_array)
corr12, _ = scst.pearsonr(energy_array,volume_array)
corrs12,v = scst.spearmanr(energy_array,volume_array)
print(covariance_12[0][1])
print(corr12)
print(corrs12)