from google.colab import drive
drive.mount('/content/drive', force_remount=True)

Mounted at /content/drive

import pandas as pd
import numpy as np
from sklearn.preprocessing import LabelBinarizer

from tensorflow import keras
originalModel = keras.models.load_model('/content/drive/MyDrive/Speech Recognition/orginalModel')
noisyModel = keras.models.load_model('/content/drive/MyDrive/Speech Recognition/noisyModel')

!pip3 install pickle5
import pickle5 as pickle
with open("/content/drive/MyDrive/Speech Recognition/mfcc.pickle", "rb") as fh:
  data = pickle.load(fh)
data.to_pickle("p4")
dfOriginal = pd.read_pickle("p4")

with open("/content/drive/MyDrive/Speech Recognition/noisy.pickle", "rb") as fh:
  data = pickle.load(fh)
data.to_pickle("p4")
dfNoisy = pd.read_pickle("p4")

Requirement already satisfied: pickle5 in /usr/local/lib/python3.7/dist-packages (0.0.11)

def data_shaper(df):
  y = df['word']
  X = df['mfcc']
  for i in range(len(X)):
    
    if X[i].shape == (32,32): 
        np.asarray(X[i]).astype(np.float32)
    else:
        a = np.empty([32,32])
        f1 = X[i].shape[0]
        f2 = X[i].shape[1]
        a[:f1,:f2]=X[i]
        X[i] = np.asarray(a).astype(np.float32)

  X = np.dstack(X).T
  print(X.shape,len(X))
  from keras.layers import Normalization
  layer = Normalization(mean=0., variance=1.)
  X = layer(X)

  lb = LabelBinarizer()
  lbFit = lb.fit_transform(y)
  lbFit.shape
  return [X, lbFit]

def test(Model, X, y):
    y_pred = Model.predict(X)
    t = 0
    c = 0 
    for i in range(len(y_pred)):
      t = t+1
      yyy = (np.max(y_pred[i])==y_pred[i])
      yP = yyy*(np.ones([1,20])[0]).astype(np.int)
      if np.array_equal(yP, y[i]):
        c = c+1
      if (i%5000)==0:
          print(c/t)
    z = c/t
    print(z)
    return z

Classification on Original Data with a Regular Model

X, y = data_shaper(dfOriginal)
test(originalModel,X,y)

(47348, 32, 32) 47348
1.0
0.9334133173365327
0.9343065693430657
0.9357376174921672
0.9354532273386331
0.9350425982960682
0.9352354921502617
0.9368303762749636
0.9378515537111572
0.9373125041665741
0.9382656078398243

0.9382656078398243

Classification on Original Data with the Noise Tolerant Model

X, y = data_shaper(dfOriginal)
test(noisyModel,X,y)

Classification on Noisy Data with the Regular Model

X, y = data_shaper(dfNoisy)
test(originalModel,X,y)

Classification on Noisy Data with the Noise Tolerant Model

X, y = data_shaper(dfNoisy)
test(noisyModel,X,y)