rcnn-2d-prolongation_train.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "from keras import backend as K\n",
    "from keras.layers import Input, Dense\n",
    "from keras.models import Model\n",
    "from keras.layers import Dense, Dropout, Reshape, Permute\n",
    "from keras.layers.convolutional import Convolution2D\n",
    "from keras.layers.convolutional import MaxPooling2D, ZeroPadding2D\n",
    "from keras.layers.normalization import BatchNormalization\n",
    "from keras.layers.advanced_activations import ReLU\n",
    "from keras.layers.recurrent import GRU\n",
    "from keras.utils.data_utils import get_file\n",
    "from keras.layers import Activation, Dense, Dropout, Conv2D, Flatten, MaxPooling2D\n",
    "import librosa\n",
    "import time\n",
    "import numpy as np\n",
    "np.set_printoptions(threshold=np.nan)\n",
    "import tensorflow as tf\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline\n",
    "plt.style.use('ggplot')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((1926, 2, 44), (1926, 2))"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_data1 = np.load('mfcc_select1_pro_bkgn.npy')\n",
    "X_data2 = np.load('mfcc_select2_pro_bkgn.npy')\n",
    "y_data1 = np.load('label_select1_pro_bkgn.npy')\n",
    "y_data2 = np.load('label_select2_pro_bkgn.npy')\n",
    "\n",
    "X_data = np.concatenate((X_data1,X_data2),axis=0)\n",
    "y_data = np.concatenate((y_data1,y_data2),axis=0)\n",
    "\n",
    "X_data.shape, y_data.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((657, 2, 44), (657, 2))"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_data = np.load('mfcc_select2_pro.npy')\n",
    "y_data = np.load('label_select2_pro.npy')\n",
    "\n",
    "X_data.shape, y_data.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[-349.38297733 -339.46299205 -335.9631354  -332.85456749 -335.33701397\n",
      " -338.99786663 -335.77088785 -333.44439033 -329.44559082 -322.16420586\n",
      " -319.92502913 -316.64271557 -311.43076845 -308.73829721 -309.3209236\n",
      " -307.67348592 -303.83712975 -305.39611913 -310.7827827  -312.72902822\n",
      " -320.60781548 -332.39978589 -345.72288923 -354.45209134 -345.82691536\n",
      " -222.58115458 -144.63731484 -104.18134994  -94.19182443 -125.13382514\n",
      " -176.27034638 -187.67775668 -142.13757249 -106.01491192 -103.24835776\n",
      " -105.74849192 -113.21542036 -124.97840797 -130.46651078 -136.70618579\n",
      " -157.23289474 -202.31283386 -221.40212359 -224.78087842]\n",
      "after\n",
      "[-0.19771582 -0.1921021  -0.19012153 -0.18836239 -0.18976721 -0.19183888\n",
      " -0.19001274 -0.18869617 -0.18643325 -0.18231272 -0.18104557 -0.17918811\n",
      " -0.17623866 -0.174715   -0.1750447  -0.17411242 -0.17194142 -0.17282366\n",
      " -0.17587197 -0.17697335 -0.18143196 -0.18810503 -0.19564457 -0.20058443\n",
      " -0.19570344 -0.12595867 -0.08185025 -0.05895622 -0.05330315 -0.07081323\n",
      " -0.09975138 -0.10620684 -0.08043565 -0.05999384 -0.05842824 -0.05984307\n",
      " -0.0640686  -0.07072528 -0.073831   -0.07736203 -0.08897809 -0.11448883\n",
      " -0.12529146 -0.12720349]\n"
     ]
    }
   ],
   "source": [
    "from sklearn.preprocessing import StandardScaler\n",
    "print(X_data[0][0])\n",
    "X_data = tf.keras.utils.normalize(X_data, axis=-1, order=2)\n",
    "#X_data = StandardScaler().fit_transform(X_data)\n",
    "print(\"after\")\n",
    "print(X_data[0][0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(1926, 2, 44)\n",
      "(1926, 2, 44, 1)\n",
      "ns = 1542\n",
      "s = 384\n"
     ]
    }
   ],
   "source": [
    "print(X_data.shape)\n",
    "X_data = np.array(X_data.reshape(X_data.shape[0], 2, 44, 1))\n",
    "print(X_data.shape)\n",
    "y=[]\n",
    "\n",
    "ns=s=0\n",
    "for i in range(len(y_data)):\n",
    "    if y_data[i][0]==1:\n",
    "        y.append('0')\n",
    "        ns+=1\n",
    "    else:\n",
    "        y.append('1')\n",
    "        s+=1\n",
    "\n",
    "print(\"ns = \"+str(ns))\n",
    "print(\"s = \"+str(s))\n",
    "y = np.array(y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.model_selection import train_test_split\n",
    "#X_sub, X_test, y_sub, y_test = train_test_split(X_data, y_data, test_size=0.2)\n",
    "X_train, X_test, y_train, y_test = train_test_split(X_data, y, test_size=0.02)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(1887, 39, 1887, 39)"
      ]
     },
     "execution_count": 45,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(X_train), len(X_test), len(y_train), len(y_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((1887, 2, 44, 1), (39, 2, 44, 1), (1887,), (39,))"
      ]
     },
     "execution_count": 46,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_train.shape, X_test.shape, y_train.shape, y_test.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "conv2d_17 (Conv2D)           (None, 2, 20, 32)         192       \n",
      "_________________________________________________________________\n",
      "activation_25 (Activation)   (None, 2, 20, 32)         0         \n",
      "_________________________________________________________________\n",
      "conv2d_18 (Conv2D)           (None, 2, 8, 32)          5152      \n",
      "_________________________________________________________________\n",
      "activation_26 (Activation)   (None, 2, 8, 32)          0         \n",
      "_________________________________________________________________\n",
      "reshape_9 (Reshape)          (None, 16, 32)            0         \n",
      "_________________________________________________________________\n",
      "gru1 (GRU)                   (None, 16, 32)            6240      \n",
      "_________________________________________________________________\n",
      "gru2 (GRU)                   (None, 32)                6240      \n",
      "_________________________________________________________________\n",
      "dropout_9 (Dropout)          (None, 32)                0         \n",
      "_________________________________________________________________\n",
      "dense_9 (Dense)              (None, 1)                 33        \n",
      "_________________________________________________________________\n",
      "activation_27 (Activation)   (None, 1)                 0         \n",
      "=================================================================\n",
      "Total params: 17,857\n",
      "Trainable params: 17,857\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n",
      "None\n"
     ]
    }
   ],
   "source": [
    "from keras.layers import Activation, Dense, Dropout, Conv2D, Flatten, MaxPooling2D\n",
    "from keras.models import Sequential\n",
    "\n",
    "model = Sequential()\n",
    "\n",
    "# Model Input\n",
    "input_shape=(2, 44, 1)\n",
    "\n",
    "# Using CNN to build model\n",
    "# W=(W−F+2P)/S+1\n",
    "# 16 depths (2-1)/1+1 x (44-5)/2+1 = 2 x 20 x 16\n",
    "model.add(Conv2D(32, (1, 5), strides=(1, 2), input_shape=input_shape))\n",
    "#model.add(BatchNormalization(axis=-1))\n",
    "model.add(Activation('relu'))\n",
    "\n",
    "# 2 x 8 x 32\n",
    "model.add(Conv2D(32, (1, 5), strides=(1, 2), padding=\"valid\"))\n",
    "#model.add(BatchNormalization(axis=-1))\n",
    "model.add(Activation('relu'))\n",
    "\n",
    "model.add(Reshape((16, 32)))\n",
    "\n",
    "# GRU block 1, 2, output\n",
    "model.add(GRU(32, return_sequences=True, name='gru1'))\n",
    "model.add(GRU(32, return_sequences=False, name='gru2'))\n",
    "#model.add(GRU(32, return_sequences=False, name='gru3'))\n",
    "model.add(Dropout(0.3))\n",
    "\n",
    "# Output\n",
    "model.add(Dense(1))\n",
    "model.add(Activation('sigmoid'))\n",
    "\n",
    "print(model.summary())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 1887 samples, validate on 39 samples\n",
      "Epoch 1/64\n",
      "1887/1887 [==============================] - 3s 2ms/step - loss: 0.5302 - acc: 0.7822 - val_loss: 0.3798 - val_acc: 0.8462\n",
      "Epoch 2/64\n",
      "1887/1887 [==============================] - 1s 756us/step - loss: 0.4845 - acc: 0.8066 - val_loss: 0.3597 - val_acc: 0.8462\n",
      "Epoch 3/64\n",
      "1887/1887 [==============================] - 1s 760us/step - loss: 0.4750 - acc: 0.8066 - val_loss: 0.3432 - val_acc: 0.8462\n",
      "Epoch 4/64\n",
      "1887/1887 [==============================] - 1s 756us/step - loss: 0.4692 - acc: 0.8124 - val_loss: 0.3657 - val_acc: 0.8462\n",
      "Epoch 5/64\n",
      "1887/1887 [==============================] - 1s 760us/step - loss: 0.4576 - acc: 0.8156 - val_loss: 0.3641 - val_acc: 0.8462\n",
      "Epoch 6/64\n",
      "1887/1887 [==============================] - 1s 763us/step - loss: 0.4449 - acc: 0.8203 - val_loss: 0.3687 - val_acc: 0.8462\n",
      "Epoch 7/64\n",
      "1887/1887 [==============================] - 1s 753us/step - loss: 0.4460 - acc: 0.8198 - val_loss: 0.3481 - val_acc: 0.8462\n",
      "Epoch 8/64\n",
      "1887/1887 [==============================] - 1s 777us/step - loss: 0.4342 - acc: 0.8272 - val_loss: 0.3714 - val_acc: 0.8462\n",
      "Epoch 9/64\n",
      "1887/1887 [==============================] - 1s 778us/step - loss: 0.4283 - acc: 0.8278 - val_loss: 0.3960 - val_acc: 0.8462\n",
      "Epoch 10/64\n",
      "1887/1887 [==============================] - 1s 773us/step - loss: 0.4295 - acc: 0.8272 - val_loss: 0.4049 - val_acc: 0.8462\n",
      "Epoch 11/64\n",
      "1887/1887 [==============================] - 1s 787us/step - loss: 0.4172 - acc: 0.8288 - val_loss: 0.4165 - val_acc: 0.8462\n",
      "Epoch 12/64\n",
      "1887/1887 [==============================] - 1s 771us/step - loss: 0.4064 - acc: 0.8336 - val_loss: 0.4069 - val_acc: 0.8462\n",
      "Epoch 13/64\n",
      "1887/1887 [==============================] - 1s 775us/step - loss: 0.3905 - acc: 0.8468 - val_loss: 0.4207 - val_acc: 0.8462\n",
      "Epoch 14/64\n",
      "1887/1887 [==============================] - 1s 787us/step - loss: 0.3855 - acc: 0.8442 - val_loss: 0.4432 - val_acc: 0.7949\n",
      "Epoch 15/64\n",
      "1887/1887 [==============================] - 2s 799us/step - loss: 0.3640 - acc: 0.8463 - val_loss: 0.4330 - val_acc: 0.8205\n",
      "Epoch 16/64\n",
      "1887/1887 [==============================] - 1s 785us/step - loss: 0.3618 - acc: 0.8484 - val_loss: 0.4755 - val_acc: 0.7949\n",
      "Epoch 17/64\n",
      "1887/1887 [==============================] - 1s 770us/step - loss: 0.3539 - acc: 0.8606 - val_loss: 0.4200 - val_acc: 0.8205\n",
      "Epoch 18/64\n",
      "1887/1887 [==============================] - 1s 785us/step - loss: 0.3475 - acc: 0.8601 - val_loss: 0.3996 - val_acc: 0.7949\n",
      "Epoch 19/64\n",
      "1887/1887 [==============================] - 2s 802us/step - loss: 0.3412 - acc: 0.8633 - val_loss: 0.4988 - val_acc: 0.7949\n",
      "Epoch 20/64\n",
      "1887/1887 [==============================] - 2s 872us/step - loss: 0.3178 - acc: 0.8733 - val_loss: 0.4718 - val_acc: 0.7949\n",
      "Epoch 21/64\n",
      "1887/1887 [==============================] - 1s 775us/step - loss: 0.3116 - acc: 0.8797 - val_loss: 0.4274 - val_acc: 0.7949\n",
      "Epoch 22/64\n",
      "1887/1887 [==============================] - 1s 794us/step - loss: 0.3013 - acc: 0.8786 - val_loss: 0.5163 - val_acc: 0.7949\n",
      "Epoch 23/64\n",
      "1887/1887 [==============================] - 1s 792us/step - loss: 0.2950 - acc: 0.8818 - val_loss: 0.6137 - val_acc: 0.7949\n",
      "Epoch 24/64\n",
      "1887/1887 [==============================] - 1s 778us/step - loss: 0.2836 - acc: 0.8861 - val_loss: 0.4688 - val_acc: 0.7949\n",
      "Epoch 25/64\n",
      "1887/1887 [==============================] - 1s 789us/step - loss: 0.2774 - acc: 0.8850 - val_loss: 0.5153 - val_acc: 0.7692\n",
      "Epoch 26/64\n",
      "1887/1887 [==============================] - 1s 786us/step - loss: 0.2680 - acc: 0.8956 - val_loss: 0.4713 - val_acc: 0.8462\n",
      "Epoch 27/64\n",
      "1887/1887 [==============================] - 2s 816us/step - loss: 0.2727 - acc: 0.8877 - val_loss: 0.5439 - val_acc: 0.7692\n",
      "Epoch 28/64\n",
      "1887/1887 [==============================] - 1s 785us/step - loss: 0.2429 - acc: 0.9062 - val_loss: 0.5462 - val_acc: 0.8205\n",
      "Epoch 29/64\n",
      "1887/1887 [==============================] - 2s 803us/step - loss: 0.2206 - acc: 0.9104 - val_loss: 0.6050 - val_acc: 0.7949\n",
      "Epoch 30/64\n",
      "1887/1887 [==============================] - 2s 809us/step - loss: 0.2336 - acc: 0.9009 - val_loss: 0.5137 - val_acc: 0.7949\n",
      "Epoch 31/64\n",
      "1887/1887 [==============================] - 1s 786us/step - loss: 0.2079 - acc: 0.9184 - val_loss: 0.6080 - val_acc: 0.7949\n",
      "Epoch 32/64\n",
      "1887/1887 [==============================] - 2s 798us/step - loss: 0.1940 - acc: 0.9269 - val_loss: 0.6026 - val_acc: 0.8205\n",
      "Epoch 33/64\n",
      "1887/1887 [==============================] - 2s 895us/step - loss: 0.2094 - acc: 0.9141 - val_loss: 0.5998 - val_acc: 0.7949\n",
      "Epoch 34/64\n",
      "1887/1887 [==============================] - 1s 757us/step - loss: 0.2026 - acc: 0.9179 - val_loss: 0.6472 - val_acc: 0.7949\n",
      "Epoch 35/64\n",
      "1887/1887 [==============================] - 1s 789us/step - loss: 0.1687 - acc: 0.9380 - val_loss: 0.6845 - val_acc: 0.7692\n",
      "Epoch 36/64\n",
      "1887/1887 [==============================] - 1s 778us/step - loss: 0.1587 - acc: 0.9401 - val_loss: 0.6455 - val_acc: 0.7949\n",
      "Epoch 37/64\n",
      "1887/1887 [==============================] - 1s 782us/step - loss: 0.1400 - acc: 0.9444 - val_loss: 0.7424 - val_acc: 0.7692\n",
      "Epoch 38/64\n",
      "1887/1887 [==============================] - 1s 790us/step - loss: 0.1115 - acc: 0.9603 - val_loss: 0.5930 - val_acc: 0.8718\n",
      "Epoch 39/64\n",
      "1887/1887 [==============================] - 1s 785us/step - loss: 0.1254 - acc: 0.9518 - val_loss: 0.6529 - val_acc: 0.8205\n",
      "Epoch 40/64\n",
      "1887/1887 [==============================] - 1s 767us/step - loss: 0.1232 - acc: 0.9518 - val_loss: 0.6755 - val_acc: 0.8205\n",
      "Epoch 41/64\n",
      "1887/1887 [==============================] - 1s 785us/step - loss: 0.1185 - acc: 0.9565 - val_loss: 0.7879 - val_acc: 0.7949\n",
      "Epoch 42/64\n",
      "1887/1887 [==============================] - 1s 777us/step - loss: 0.1141 - acc: 0.9550 - val_loss: 0.5869 - val_acc: 0.8205\n",
      "Epoch 43/64\n",
      "1887/1887 [==============================] - 1s 785us/step - loss: 0.0885 - acc: 0.9661 - val_loss: 0.8776 - val_acc: 0.7949\n",
      "Epoch 44/64\n",
      "1887/1887 [==============================] - 1s 769us/step - loss: 0.1016 - acc: 0.9571 - val_loss: 0.8920 - val_acc: 0.7692\n",
      "Epoch 45/64\n",
      "1887/1887 [==============================] - 2s 799us/step - loss: 0.0842 - acc: 0.9682 - val_loss: 0.8733 - val_acc: 0.8205\n",
      "Epoch 46/64\n",
      "1887/1887 [==============================] - 2s 795us/step - loss: 0.1033 - acc: 0.9608 - val_loss: 0.8778 - val_acc: 0.8205\n",
      "Epoch 47/64\n",
      "1887/1887 [==============================] - 1s 789us/step - loss: 0.0691 - acc: 0.9751 - val_loss: 1.0409 - val_acc: 0.7436\n",
      "Epoch 48/64\n",
      "1887/1887 [==============================] - 2s 799us/step - loss: 0.0588 - acc: 0.9777 - val_loss: 0.9422 - val_acc: 0.7692\n",
      "Epoch 49/64\n",
      "1887/1887 [==============================] - 1s 774us/step - loss: 0.0483 - acc: 0.9841 - val_loss: 0.7784 - val_acc: 0.7949\n",
      "Epoch 50/64\n",
      "1887/1887 [==============================] - 1s 788us/step - loss: 0.0687 - acc: 0.9772 - val_loss: 0.8758 - val_acc: 0.7179\n",
      "Epoch 51/64\n",
      "1887/1887 [==============================] - 1s 794us/step - loss: 0.0479 - acc: 0.9862 - val_loss: 1.0529 - val_acc: 0.8205\n",
      "Epoch 52/64\n",
      "1887/1887 [==============================] - 1s 787us/step - loss: 0.0325 - acc: 0.9894 - val_loss: 0.9509 - val_acc: 0.8205\n",
      "Epoch 53/64\n",
      "1887/1887 [==============================] - 1s 779us/step - loss: 0.0200 - acc: 0.9942 - val_loss: 0.8637 - val_acc: 0.7949\n",
      "Epoch 54/64\n",
      "1887/1887 [==============================] - 1s 777us/step - loss: 0.0688 - acc: 0.9735 - val_loss: 0.8229 - val_acc: 0.7692\n",
      "Epoch 55/64\n",
      "1887/1887 [==============================] - 2s 798us/step - loss: 0.0859 - acc: 0.9698 - val_loss: 1.0391 - val_acc: 0.7436\n",
      "Epoch 56/64\n",
      "1887/1887 [==============================] - 2s 804us/step - loss: 0.0927 - acc: 0.9650 - val_loss: 0.9856 - val_acc: 0.7692\n",
      "Epoch 57/64\n",
      "1887/1887 [==============================] - 1s 793us/step - loss: 0.0496 - acc: 0.9836 - val_loss: 0.9975 - val_acc: 0.7436\n",
      "Epoch 58/64\n",
      "1887/1887 [==============================] - 1s 790us/step - loss: 0.0262 - acc: 0.9942 - val_loss: 0.8314 - val_acc: 0.7692\n",
      "Epoch 59/64\n",
      "1887/1887 [==============================] - 2s 853us/step - loss: 0.0201 - acc: 0.9936 - val_loss: 1.0074 - val_acc: 0.7949\n",
      "Epoch 60/64\n",
      "1887/1887 [==============================] - 1s 774us/step - loss: 0.0206 - acc: 0.9936 - val_loss: 1.1698 - val_acc: 0.7179\n",
      "Epoch 61/64\n",
      "1887/1887 [==============================] - 1s 763us/step - loss: 0.0179 - acc: 0.9936 - val_loss: 1.2115 - val_acc: 0.7436\n",
      "Epoch 62/64\n",
      "1887/1887 [==============================] - 1s 749us/step - loss: 0.0432 - acc: 0.9883 - val_loss: 0.9619 - val_acc: 0.7179\n",
      "Epoch 63/64\n",
      "1887/1887 [==============================] - 1s 739us/step - loss: 0.0599 - acc: 0.9777 - val_loss: 1.0045 - val_acc: 0.8205\n",
      "Epoch 64/64\n",
      "1887/1887 [==============================] - 1s 749us/step - loss: 0.0473 - acc: 0.9793 - val_loss: 1.1592 - val_acc: 0.8205\n",
      "39/39 [==============================] - 0s 1ms/step\n",
      "Test loss: 1.15924214553\n",
      "Test accuracy: 0.820512820513\n"
     ]
    }
   ],
   "source": [
    "model.compile(\n",
    "    optimizer=\"Adam\",\n",
    "    loss=\"binary_crossentropy\",\n",
    "    metrics=['accuracy'])\n",
    "\n",
    "model.fit(\n",
    "    x=X_train,\n",
    "    y=y_train,\n",
    "    epochs=64,\n",
    "    batch_size=64,\n",
    "    validation_data= (X_test, y_test),\n",
    ")\n",
    "\n",
    "score = model.evaluate(x=X_test,y=y_test)\n",
    "\n",
    "print('Test loss:', score[0])\n",
    "print('Test accuracy:', score[1])\n",
    "\n",
    "#timestr = time.strftime('%Y%m%d-%H%M%S')\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [],
   "source": [
    "modelName = 'rcnn-2d-select1and2-pro-exns-bkgn-1.h5'.format(timestr)\n",
    "model.save('models/{}'.format(modelName))\n",
    "print('Model exported and finished')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "ff\n",
      "[]\n",
      "(133, 2, 44)\n",
      "chunk 1 --> [[ 0.99000001]]\n",
      "chunk 2 --> [[ 0.99000001]]\n",
      "chunk 3 --> [[ 0.97000003]]\n",
      "chunk 4 --> [[ 0.82999998]]\n",
      "chunk 5 --> [[ 0.99000001]]\n",
      "chunk 6 --> [[ 0.98000002]]\n",
      "chunk 7 --> [[ 0.56]]\n",
      "chunk 8 --> [[ 0.99000001]]\n",
      "chunk 9 --> [[ 0.99000001]]\n",
      "chunk 10 --> [[ 0.99000001]]\n",
      "chunk 11 --> [[ 0.99000001]]\n",
      "chunk 12 --> [[ 0.98000002]]\n",
      "chunk 13 --> [[ 0.99000001]]\n",
      "chunk 14 --> [[ 0.99000001]]\n",
      "chunk 15 --> [[ 0.97000003]]\n",
      "chunk 16 --> [[ 0.97000003]]\n",
      "chunk 17 --> [[ 0.99000001]]\n",
      "chunk 18 --> [[ 0.99000001]]\n",
      "chunk 19 --> [[ 0.99000001]]\n",
      "chunk 20 --> [[ 0.99000001]]\n",
      "chunk 21 --> [[ 0.99000001]]\n",
      "chunk 22 --> [[ 0.97000003]]\n",
      "chunk 23 --> [[ 0.99000001]]\n",
      "chunk 24 --> [[ 0.99000001]]\n",
      "chunk 25 --> [[ 0.98000002]]\n",
      "chunk 26 --> [[ 0.99000001]]\n",
      "chunk 27 --> [[ 0.99000001]]\n",
      "chunk 28 --> [[ 0.99000001]]\n",
      "chunk 29 --> [[ 0.99000001]]\n",
      "chunk 30 --> [[ 0.98000002]]\n",
      "chunk 31 --> [[ 0.99000001]]\n",
      "chunk 32 --> [[ 0.99000001]]\n",
      "chunk 33 --> [[ 0.99000001]]\n",
      "chunk 34 --> [[ 0.99000001]]\n",
      "chunk 35 --> [[ 0.99000001]]\n",
      "chunk 36 --> [[ 0.99000001]]\n",
      "chunk 37 --> [[ 0.99000001]]\n",
      "chunk 38 --> [[ 0.99000001]]\n",
      "chunk 39 --> [[ 0.99000001]]\n",
      "chunk 40 --> [[ 1.]]\n",
      "chunk 41 --> [[ 0.99000001]]\n",
      "chunk 42 --> [[ 0.99000001]]\n",
      "chunk 43 --> [[ 0.99000001]]\n",
      "chunk 44 --> [[ 0.99000001]]\n",
      "chunk 45 --> [[ 0.97000003]]\n",
      "chunk 46 --> [[ 0.99000001]]\n",
      "chunk 47 --> [[ 0.97000003]]\n",
      "chunk 48 --> [[ 1.]]\n",
      "chunk 49 --> [[ 0.99000001]]\n",
      "chunk 50 --> [[ 0.99000001]]\n",
      "chunk 51 --> [[ 0.91000003]]\n",
      "chunk 52 --> [[ 1.]]\n",
      "chunk 53 --> [[ 0.99000001]]\n",
      "chunk 54 --> [[ 0.99000001]]\n",
      "chunk 55 --> [[ 1.]]\n",
      "chunk 56 --> [[ 0.81]]\n",
      "chunk 57 --> [[ 0.99000001]]\n",
      "chunk 58 --> [[ 0.99000001]]\n",
      "chunk 59 --> [[ 0.99000001]]\n",
      "chunk 60 --> [[ 0.69]]\n",
      "chunk 61 --> [[ 0.86000001]]\n",
      "chunk 62 --> [[ 0.99000001]]\n",
      "chunk 63 --> [[ 0.57999998]]\n",
      "chunk 64 --> [[ 0.99000001]]\n",
      "chunk 65 --> [[ 0.99000001]]\n",
      "chunk 66 --> [[ 0.99000001]]\n",
      "chunk 67 --> [[ 0.30000001]]\n",
      "chunk 68 --> [[ 0.99000001]]\n",
      "chunk 69 --> [[ 0.89999998]]\n",
      "chunk 70 --> [[ 0.99000001]]\n",
      "chunk 71 --> [[ 0.99000001]]\n",
      "chunk 72 --> [[ 0.99000001]]\n",
      "chunk 73 --> [[ 0.97000003]]\n",
      "chunk 74 --> [[ 0.99000001]]\n",
      "chunk 75 --> [[ 0.99000001]]\n",
      "chunk 76 --> [[ 0.99000001]]\n",
      "chunk 77 --> [[ 0.98000002]]\n",
      "chunk 78 --> [[ 0.99000001]]\n",
      "chunk 79 --> [[ 0.99000001]]\n",
      "chunk 80 --> [[ 0.99000001]]\n",
      "chunk 81 --> [[ 0.99000001]]\n",
      "chunk 82 --> [[ 0.99000001]]\n",
      "chunk 83 --> [[ 0.99000001]]\n",
      "chunk 84 --> [[ 0.99000001]]\n",
      "chunk 85 --> [[ 0.99000001]]\n",
      "chunk 86 --> [[ 0.95999998]]\n",
      "chunk 87 --> [[ 0.99000001]]\n",
      "chunk 88 --> [[ 0.99000001]]\n",
      "chunk 89 --> [[ 0.99000001]]\n",
      "chunk 90 --> [[ 0.97000003]]\n",
      "chunk 91 --> [[ 0.97000003]]\n",
      "chunk 92 --> [[ 0.66000003]]\n",
      "chunk 93 --> [[ 0.99000001]]\n",
      "chunk 94 --> [[ 0.99000001]]\n",
      "chunk 95 --> [[ 0.99000001]]\n",
      "chunk 96 --> [[ 0.99000001]]\n",
      "chunk 97 --> [[ 0.99000001]]\n",
      "chunk 98 --> [[ 0.99000001]]\n",
      "chunk 99 --> [[ 0.99000001]]\n",
      "chunk 100 --> [[ 0.99000001]]\n",
      "chunk 101 --> [[ 1.]]\n",
      "chunk 102 --> [[ 0.98000002]]\n",
      "chunk 103 --> [[ 0.99000001]]\n",
      "chunk 104 --> [[ 0.99000001]]\n",
      "chunk 105 --> [[ 0.98000002]]\n",
      "chunk 106 --> [[ 0.99000001]]\n",
      "chunk 107 --> [[ 0.99000001]]\n",
      "chunk 108 --> [[ 0.99000001]]\n",
      "chunk 109 --> [[ 0.99000001]]\n",
      "chunk 110 --> [[ 0.99000001]]\n",
      "chunk 111 --> [[ 0.99000001]]\n",
      "chunk 112 --> [[ 0.99000001]]\n",
      "chunk 113 --> [[ 0.99000001]]\n",
      "chunk 114 --> [[ 0.99000001]]\n",
      "chunk 115 --> [[ 0.97000003]]\n",
      "chunk 116 --> [[ 0.99000001]]\n",
      "chunk 117 --> [[ 0.99000001]]\n",
      "chunk 118 --> [[ 0.99000001]]\n",
      "chunk 119 --> [[ 0.99000001]]\n",
      "chunk 120 --> [[ 1.]]\n",
      "chunk 121 --> [[ 0.99000001]]\n",
      "chunk 122 --> [[ 1.]]\n",
      "chunk 123 --> [[ 0.98000002]]\n",
      "chunk 124 --> [[ 0.99000001]]\n",
      "chunk 125 --> [[ 0.99000001]]\n",
      "chunk 126 --> [[ 0.99000001]]\n",
      "chunk 127 --> [[ 0.99000001]]\n",
      "chunk 128 --> [[ 0.37]]\n",
      "chunk 129 --> [[ 1.]]\n",
      "chunk 130 --> [[ 0.99000001]]\n",
      "chunk 131 --> [[ 0.99000001]]\n",
      "chunk 132 --> [[ 0.99000001]]\n",
      "chunk 133 --> [[ 0.99000001]]\n",
      "6.015037593984962\n",
      "8\n",
      "133\n",
      "96.2556401142\n"
     ]
    }
   ],
   "source": [
    "import librosa\n",
    "#import keras\n",
    "import numpy as np\n",
    "import tensorflow as tf\n",
    "from pydub import AudioSegment\n",
    "#from pydub.silence import split_on_silence\n",
    "#import pandas as pd\n",
    "import os\n",
    "from sklearn.preprocessing import StandardScaler\n",
    "from keras.models import load_model\n",
    "#import matplotlib.pyplot as plt\n",
    "#%matplotlib inline\n",
    "#plt.style.use('ggplot')\n",
    "np.set_printoptions(threshold=np.nan)\n",
    "\n",
    "sound_file = AudioSegment.from_wav(\"/home/mansi/anaconda3/beproject/stutter_det/read1s2dmf/M_0048_11y1m_2/M_0048_11y1m_2.wav\")\n",
    "\n",
    "audio_chunks = sound_file[::1000]\n",
    "\n",
    "mfcc_array = []\n",
    "print(\"ff\")\n",
    "print(mfcc_array)\n",
    "for i, chunk in enumerate(audio_chunks):\n",
    "    chunkfile = \"chunks_test/chunk{0}.wav\".format(i)\n",
    "    chunk.export(chunkfile, format=\"wav\")\n",
    "    y, sr = librosa.load(chunkfile)\n",
    "    mfcc = np.array(librosa.feature.mfcc(y=y, sr=sr, n_mfcc=13))\n",
    "    #print(mfcc.shape)\n",
    "    #mfcc_array.append(librosa.feature.mfcc(y=y, sr=sr, n_mfcc=39))\n",
    "    if mfcc.shape[0]==13 and mfcc.shape[1]==44:\n",
    "        a = []\n",
    "        a.append(mfcc[0])\n",
    "        a.append(mfcc[12])\n",
    "        mfcc_array.append(a)\n",
    "    #mfcc_array.append(np.pad(mfcc, (0,-mfcc.shape[0]), 'constant', constant_values=(0)))\n",
    "    \n",
    "mfcc_array = np.array(mfcc_array)\n",
    "\n",
    "print(mfcc_array.shape)\n",
    "\n",
    "#mfcc_array = tf.keras.utils.normalize(mfcc_array, order=2)\n",
    "#mfcc_array = StandardScaler().fit_transform(mfcc_array)\n",
    "\n",
    "z = mfcc_array.reshape(mfcc_array.shape[0], 2, 44, 1)\n",
    "\n",
    "s=0\n",
    "i=0\n",
    "t=0\n",
    "\n",
    "for f in z:\n",
    "    #pc = model.predict_classes(f.reshape(1,39,44,1), batch_size=1)\n",
    "    pp = model.predict(f.reshape(1,2,44,1), batch_size=1)\n",
    "    pp = np.around(pp,decimals=2)\n",
    "    #print(\"chunk\",str(i),\"-->\",pc,\"-->\",pp)\n",
    "    if pp[0][0] == 1.0:\n",
    "        s+=1\n",
    "    t+=pp[0][0]\n",
    "    i+=1\n",
    "    print(\"chunk\",str(i),\"-->\",pp)\n",
    "\n",
    "print(s/len(z)*100)\n",
    "print(s)\n",
    "print(len(z))\n",
    "print(t/len(z)*100)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "conv2d_9 (Conv2D)            (None, 2, 20, 32)         192       \n",
      "_________________________________________________________________\n",
      "activation_13 (Activation)   (None, 2, 20, 32)         0         \n",
      "_________________________________________________________________\n",
      "conv2d_10 (Conv2D)           (None, 2, 8, 32)          5152      \n",
      "_________________________________________________________________\n",
      "activation_14 (Activation)   (None, 2, 8, 32)          0         \n",
      "_________________________________________________________________\n",
      "reshape_5 (Reshape)          (None, 16, 32)            0         \n",
      "_________________________________________________________________\n",
      "gru1 (GRU)                   (None, 16, 32)            6240      \n",
      "_________________________________________________________________\n",
      "gru2 (GRU)                   (None, 32)                6240      \n",
      "_________________________________________________________________\n",
      "dropout_5 (Dropout)          (None, 32)                0         \n",
      "_________________________________________________________________\n",
      "dense_5 (Dense)              (None, 1)                 33        \n",
      "_________________________________________________________________\n",
      "activation_15 (Activation)   (None, 1)                 0         \n",
      "=================================================================\n",
      "Total params: 17,857\n",
      "Trainable params: 17,857\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "from keras.models import load_model\n",
    "model1 = load_model('/home/mansi/anaconda3/beproject/stutter_det/models/best_model_pro.h5')\n",
    "model1.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "M_1206_12y3m_1\n",
      "18.6214954218\n",
      "19.5046730009\n",
      "M_1202_10y7m_1\n",
      "18.4098360968\n",
      "20.9016393687\n",
      "M_1023_11y6m_1\n",
      "8.55371897871\n",
      "8.84297517367\n",
      "M_1022_10y9m_2\n",
      "26.1653543457\n",
      "27.3779527541\n",
      "M_1213_11y3m_1\n",
      "39.5792350417\n",
      "41.8743169984\n",
      "M_0253_14y6m_1\n",
      "44.3285025548\n",
      "45.1980677744\n",
      "M_0104_15y4m_1\n",
      "51.7256100003\n",
      "52.6402441489\n",
      "M_1022_10y10m_1\n",
      "19.6504065176\n",
      "20.4552845742\n",
      "M_0880_15y2m_1\n",
      "28.1390728935\n",
      "30.013245095\n",
      "M_0760_12y3m_1\n",
      "61.7670457133\n",
      "64.9886366594\n",
      "M_1215_11y0m_1\n",
      "35.7157895753\n",
      "37.6157895672\n",
      "M_1214_11y8m_1\n",
      "27.6562500373\n",
      "28.9375000133\n",
      "M_1207_13y5m_1\n",
      "12.7213114923\n",
      "13.7978142188\n",
      "M_0760_15y3m_1\n",
      "71.1739132\n",
      "72.5341616426\n",
      "M_0760_12y3m_1\n",
      "61.7670457133\n",
      "64.9886366594\n",
      "M_1023_10y10m_1\n",
      "18.1755726119\n",
      "20.0534351853\n",
      "M_0545_15y3m_1\n",
      "46.18300667\n",
      "47.8823530762\n",
      "M_0541_12y3m_1\n",
      "27.1804511995\n",
      "28.7518797862\n",
      "M_0104_18y1m_1\n",
      "14.0800000032\n",
      "16.5866666784\n",
      "M_1216_11y1m_1\n",
      "31.9658120473\n",
      "35.2991453921\n",
      "M_1205_13y1m_1\n",
      "22.1707317403\n",
      "23.5284553311\n",
      "M_1040_12y6m_1\n",
      "22.9268293071\n",
      "24.2439024696\n",
      "M_0880_14y6m_2\n",
      "13.8846154396\n",
      "14.38461545\n"
     ]
    }
   ],
   "source": [
    "import librosa\n",
    "#import keras\n",
    "import numpy as np\n",
    "import tensorflow as tf\n",
    "from pydub import AudioSegment\n",
    "#from pydub.silence import split_on_silence\n",
    "#import pandas as pd\n",
    "import os\n",
    "from sklearn.preprocessing import StandardScaler\n",
    "np.set_printoptions(threshold=np.nan)\n",
    "\n",
    "arr = ['M_0760_15y3m_1',\n",
    "       'M_1202_10y7m_1',\n",
    "       'M_1215_11y0m_1',\n",
    "       'M_1202_10y11m_1',\n",
    "       'M_1202_10y7m_1',\n",
    "       'M_0760_12y3m_1',\n",
    "       'M_0077_11y2m_1',\n",
    "       'M_0061_18y5m_2',\n",
    "       'M_0061_18y5m_1',\n",
    "       'F_1211_11y3m_1',\n",
    "       'F_0101_14y8m_1',\n",
    "       'M_1202_10y11m_1',\n",
    "       'M_1216_11y1m_1',\n",
    "       'M_0048_11y1m_2',\n",
    "       'M_0048_11y5m_1',\n",
    "       'M_0048_11y6m_1',\n",
    "       'M_0053_10y8m_1',\n",
    "       'M_0064_11y11m_1',\n",
    "       'M_0064_11y5m_1'\n",
    "      ]\n",
    "\n",
    "arr2 = [\"M_1206_12y3m_1\",\n",
    "\"M_1202_10y7m_1\",\n",
    "\"M_1023_11y6m_1\",\n",
    "\"M_1022_10y9m_2\",\n",
    "\"M_1213_11y3m_1\",\n",
    "\"M_0253_14y6m_1\",\n",
    "\"M_0104_15y4m_1\",\n",
    "\"M_1022_10y10m_1\",\n",
    "\"M_0880_15y2m_1\",\n",
    "\"M_0760_12y3m_1\",\n",
    "\"M_1215_11y0m_1\",\n",
    "\"M_1214_11y8m_1\",\n",
    "\"M_1207_13y5m_1\",\n",
    "\"M_0760_15y3m_1\",\n",
    "\"M_0760_12y3m_1\",\n",
    "\"M_1023_10y10m_1\",\n",
    "\"M_0545_15y3m_1\",\n",
    "\"M_0541_12y3m_1\",\n",
    "\"M_0104_18y1m_1\",\n",
    "\"M_1216_11y1m_1\",\n",
    "\"M_1205_13y1m_1\",\n",
    "\"M_1040_12y6m_1\",\n",
    "\"M_0880_14y6m_2\"]\n",
    "\n",
    "files = os.listdir(\"recordings\")\n",
    "\n",
    "for audio in arr2 :\n",
    "    #if audio.endswith('.wav'):\n",
    "    print(audio)\n",
    "    #print(str(i)+'.wav')\n",
    "    #sound_file = AudioSegment.from_wav(\"/home/mansi/anaconda3/beproject/stutter_det/recordings/\"+audio)\n",
    "    sound_file = AudioSegment.from_wav(\"/home/mansi/anaconda3/beproject/stutter_det/test_rep/\"+audio+'.wav')\n",
    "    audio_chunks = sound_file[::1000]\n",
    "\n",
    "    mfcc_array = []\n",
    "\n",
    "    for i, chunk in enumerate(audio_chunks):\n",
    "        chunkfile = \"chunks_test/chunk{0}.wav\".format(i)\n",
    "        chunk.export(chunkfile, format=\"wav\")\n",
    "        y, sr = librosa.load(chunkfile)\n",
    "        mfcc = np.array(librosa.feature.mfcc(y=y, sr=sr, n_mfcc=13))\n",
    "        if mfcc.shape[0]==13 and mfcc.shape[1]==44:\n",
    "            a = []\n",
    "            a.append(mfcc[0])\n",
    "            a.append(mfcc[12])\n",
    "            mfcc_array.append(a)\n",
    "\n",
    "    mfcc_array = np.array(mfcc_array)\n",
    "    #print(mfcc_array.shape)\n",
    "    \n",
    "    z = mfcc_array.reshape(mfcc_array.shape[0], 2, 44, 1)\n",
    "    s=0\n",
    "    i=0\n",
    "    t=0\n",
    "    for f in z:\n",
    "        pp = model1.predict(f.reshape(1,2,44,1), batch_size=1)\n",
    "        pp = np.around(pp,decimals=2)\n",
    "        \n",
    "        s+=pp[0][0]\n",
    "        if pp[0][0] > 0.5:\n",
    "            t+=pp[0][0]\n",
    "        #print(\"chunk\",str(i),\"-->\",pp)\n",
    "        i+=1\n",
    "    print(t/len(z)*100)\n",
    "    print(s/len(z)*100)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}