Speaker_Classification_mfcc&gfcc_v1_10_100epoch.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "language_info": {
      "codemirror_mode": {
        "name": "ipython",
        "version": 3
      },
      "file_extension": ".py",
      "mimetype": "text/x-python",
      "name": "python",
      "nbconvert_exporter": "python",
      "pygments_lexer": "ipython3",
      "version": "3.6.9"
    },
    "colab": {
      "name": "Speaker_Classification_mfcc&gfcc_v1_10_100epoch.ipynb",
      "provenance": [],
      "collapsed_sections": [
        "qlUm3WDCPrt_"
      ],
      "include_colab_link": true
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/ritikraj660/Speaker-recognition/blob/master/Speaker_Classification_mfcc%26gfcc_v1_10_100epoch.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "CpNLfJBO1-WH",
        "colab_type": "text"
      },
      "source": [
        "# Installing Packages:"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "wj01p9v5K199",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#!pip install --upgrade tensorflow-gpu"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "9RXw9shFmGBg",
        "colab_type": "code",
        "outputId": "74c32d3e-9fec-4d7a-b2c6-68bbdb9da943",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 258
        }
      },
      "source": [
        "!pip install soundfile\n",
        "!pip install spafe"
      ],
      "execution_count": 2,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Collecting soundfile\n",
            "  Downloading https://files.pythonhosted.org/packages/eb/f2/3cbbbf3b96fb9fa91582c438b574cff3f45b29c772f94c400e2c99ef5db9/SoundFile-0.10.3.post1-py2.py3-none-any.whl\n",
            "Requirement already satisfied: cffi>=1.0 in /usr/local/lib/python3.6/dist-packages (from soundfile) (1.14.0)\n",
            "Requirement already satisfied: pycparser in /usr/local/lib/python3.6/dist-packages (from cffi>=1.0->soundfile) (2.20)\n",
            "Installing collected packages: soundfile\n",
            "Successfully installed soundfile-0.10.3.post1\n",
            "Collecting spafe\n",
            "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/a9/d9/d92807657a5125318639aa1636c0a76fb87222d48b5c290996ed9d5be1b6/spafe-0.1.2-py3-none-any.whl (77kB)\n",
            "\u001b[K     |████████████████████████████████| 81kB 3.9MB/s \n",
            "\u001b[?25hRequirement already satisfied: scipy>=1.3.1 in /usr/local/lib/python3.6/dist-packages (from spafe) (1.4.1)\n",
            "Requirement already satisfied: numpy>=1.17.2 in /usr/local/lib/python3.6/dist-packages (from spafe) (1.18.4)\n",
            "Installing collected packages: spafe\n",
            "Successfully installed spafe-0.1.2\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "rBgxeaV6jtiB",
        "colab_type": "code",
        "outputId": "0b12e87b-4b4c-41f4-cfee-cbc20e215d00",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "source": [
        "import tensorflow\n",
        "tensorflow.test.gpu_device_name() "
      ],
      "execution_count": 3,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "'/device:GPU:0'"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 3
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "x0XYclYb2NlS",
        "colab_type": "text"
      },
      "source": [
        "# Import Data from Drive :"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "o7LFfT5dOyO6",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "from pydrive.auth import GoogleAuth\n",
        "from pydrive.drive import GoogleDrive\n",
        "from google.colab import auth\n",
        "from oauth2client.client import GoogleCredentials"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ulBEeHGOPVb5",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "auth.authenticate_user()\n",
        "gauth = GoogleAuth()\n",
        "gauth.credentials = GoogleCredentials.get_application_default()\n",
        "drive = GoogleDrive(gauth)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "bFtoL6s2PVFJ",
        "colab_type": "code",
        "outputId": "86dfd4b7-525a-43ec-d9fb-789ef2467210",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "source": [
        "link = 'https://drive.google.com/file/d/1qxZWgx3vttou1HSvyxRDiOYWnaLAK9fj/view?usp=sharing'\n",
        "#fluff, id = link.split('=')\n",
        "id = '1qxZWgx3vttou1HSvyxRDiOYWnaLAK9fj'\n",
        "print (id) "
      ],
      "execution_count": 6,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "1qxZWgx3vttou1HSvyxRDiOYWnaLAK9fj\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "GpzvACTgPyj0",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "import pandas as pd\n",
        "downloaded = drive.CreateFile({'id':id}) \n",
        "downloaded.GetContentFile('LibriSpeech.zip')  "
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "MCRUyoks2WwX",
        "colab_type": "text"
      },
      "source": [
        "# Extracting Data from (.)rar"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "PUsfP_J5PySI",
        "colab_type": "code",
        "outputId": "c5e54b94-56ef-4fca-de78-f5df8b1c0f63",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 102
        }
      },
      "source": [
        "!pip install patool"
      ],
      "execution_count": 8,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Collecting patool\n",
            "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/43/94/52243ddff508780dd2d8110964320ab4851134a55ab102285b46e740f76a/patool-1.12-py2.py3-none-any.whl (77kB)\n",
            "\r\u001b[K     |████▎                           | 10kB 16.9MB/s eta 0:00:01\r\u001b[K     |████████▌                       | 20kB 4.4MB/s eta 0:00:01\r\u001b[K     |████████████▊                   | 30kB 5.6MB/s eta 0:00:01\r\u001b[K     |█████████████████               | 40kB 5.3MB/s eta 0:00:01\r\u001b[K     |█████████████████████▏          | 51kB 5.0MB/s eta 0:00:01\r\u001b[K     |█████████████████████████▍      | 61kB 5.5MB/s eta 0:00:01\r\u001b[K     |█████████████████████████████▋  | 71kB 6.0MB/s eta 0:00:01\r\u001b[K     |████████████████████████████████| 81kB 3.8MB/s \n",
            "\u001b[?25hInstalling collected packages: patool\n",
            "Successfully installed patool-1.12\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ffyDpxaCSgDx",
        "colab_type": "code",
        "outputId": "4ac221b2-05f9-4311-bb3b-bfede87ea0a9",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "source": [
        "import os\n",
        "os.listdir()"
      ],
      "execution_count": 9,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "['.config', 'adc.json', 'LibriSpeech.zip', 'sample_data']"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 9
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "CH0Ug5BPShB8",
        "colab_type": "code",
        "outputId": "ece3b300-fd98-47bd-a2ae-672142749f01",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 102
        }
      },
      "source": [
        "%%time\n",
        "import patoolib\n",
        "patoolib.extract_archive(\"LibriSpeech.zip\")"
      ],
      "execution_count": 10,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "patool: Extracting LibriSpeech.zip ...\n",
            "patool: running /usr/bin/7z x -o./Unpack_x0pv7dbc -- LibriSpeech.zip\n",
            "patool: ... LibriSpeech.zip extracted to `LibriSpeech'.\n",
            "CPU times: user 11.5 ms, sys: 17.1 ms, total: 28.5 ms\n",
            "Wall time: 603 ms\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "fcQzszcy2htw",
        "colab_type": "text"
      },
      "source": [
        "# Audio Data Preprocessing :"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Von9Wv-VP_1g",
        "colab_type": "code",
        "outputId": "03fa6dfb-e8c2-451c-b599-e3bbdc3716c3",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "source": [
        "import librosa\n",
        "import soundfile\n",
        "import tqdm\n",
        "from scipy.io import wavfile\n",
        "import glob\n",
        "import os,pickle\n",
        "import numpy as np\n",
        "from spafe.features.mfcc import mfcc, imfcc\n",
        "import csv\n",
        "import pandas as pd\n",
        "from tqdm import tqdm\n",
        "from sklearn.model_selection import train_test_split\n",
        "import numpy as np\n",
        "import tensorflow.python.keras\n",
        "from tensorflow.python.keras.models import Sequential\n",
        "from tensorflow.python.keras.layers import Dense, Dropout, Flatten, BatchNormalization\n",
        "from tensorflow.python.keras.layers import Conv2D, MaxPooling2D\n",
        "from keras.optimizers import SGD"
      ],
      "execution_count": 11,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Using TensorFlow backend.\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "73C5dhr-me2y",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def norm_(x,mean = False):\n",
        "    if mean == True:\n",
        "        return (x - np.mean(x)) / (np.max(x) - np.min(x)) #normalization                             \n",
        "    else:\n",
        "        return (x-np.mean(x)/np.std(x)) #standardization\n",
        "\n",
        "def zero_handling(x):\n",
        "    \"\"\"\n",
        "    handle the issue with zero values if they are exposed to become an argument\n",
        "    for any log function.\n",
        "\n",
        "    Returns:\n",
        "        vector with zeros substituted with epsilon values.\n",
        "    \"\"\"\n",
        "    return np.where(x == float(0), np.finfo(float).eps, x)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "mSPPghULme23",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "from spafe.features.lpc import lpc\n",
        "from spafe.features.gfcc import gfcc\n",
        "\n",
        "def mfcc_(sig,fs,cep):\n",
        "    mfccs  = mfcc(sig, fs=fs, num_ceps=cep, pre_emph=1, pre_emph_coeff=0.95, \n",
        "              win_len=0.025, win_hop=0.01, win_type='hamming', nfilts=26, \n",
        "              nfft=512, low_freq=None, high_freq=None, scale='constant', \n",
        "              dct_type=2, use_energy=False, lifter=22)\n",
        "    #print(fs)\n",
        "    return mfccs\n",
        "\n",
        "def gfcc_(sig,fs,cep):\n",
        "  gfccs = gfcc(sig, fs=fs, num_ceps=cep, pre_emph=0, \n",
        "             pre_emph_coeff=0.97, win_len=0.025, win_hop=0.01, win_type='hamming', \n",
        "             nfilts=26, nfft=512, low_freq=None, high_freq=None, scale='constant', \n",
        "             dct_type=2, use_energy=False, lifter=22, normalize=1)\n",
        "  return gfccs\n",
        "\n",
        "# combine mfcc and gfcc ceptral features\n",
        "def gfcc_mfcc(y,sr,cep):\n",
        "    g = gfcc_(y,sr,cep)\n",
        "    m = mfcc_(y,sr,cep)\n",
        "    if g.shape[0] != m.shape[0]:\n",
        "        pass\n",
        "    result = np.zeros([g.shape[1]+m.shape[1]])\n",
        "    for i in range(g.shape[0]):\n",
        "        merge = np.concatenate((m[i],g[i]))\n",
        "        result = np.vstack((result,merge))\n",
        "    return result[1:,:]"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "xaE6275lme26",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def label(user,total): # its creating lable for splited audio data\n",
        "    k_train = []\n",
        "    for i in range(total):\n",
        "        k_train.append(user)\n",
        "    y_train=np.array(k_train).T\n",
        "    return y_train"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "aV0iqD15me3D",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "from pathlib import Path\n",
        "len_audio = 148\n",
        "def s2c(mfcc=False,gmfcc=False,gfcc=False,cep=20):\n",
        "    x,y=[],[]\n",
        "    if gmfcc==True:\n",
        "      ceps=cep+cep\n",
        "    else:\n",
        "      ceps=cep  \n",
        "    data = np.empty([1,len_audio,ceps])\n",
        "    data_label = np.asarray([])\n",
        "    count =0  # E:\\dataset\\LibriSpeech\\dev-clean\n",
        "    #for file in tqdm(glob.iglob('/LibriSpeech/test-clean//**/*.flac', recursive=True)):\n",
        "    for file in tqdm(Path('LibriSpeech/dev-clean').glob('**/*.flac')):\n",
        "        file_name=os.path.basename(file)\n",
        "        user = file_name.split(\"-\")[0]\n",
        "        #print(f)\n",
        "        with soundfile.SoundFile(file) as sound_file:\n",
        "                signal = sound_file.read(dtype=\"float32\")\n",
        "                sr=sound_file.samplerate\n",
        "        signal = norm_(signal,mean = False)\n",
        "        signal = zero_handling(signal)\n",
        "        \n",
        "        if mfcc:\n",
        "            feature = mfcc_(signal,sr,cep)\n",
        "            #print(feature.shape[0])\n",
        "        if gmfcc:\n",
        "            feature = gfcc_mfcc(signal,sr,cep)\n",
        "            ceps = feature.shape[1]   \n",
        "        #if feature.shape[0]>256:\n",
        "        if gfcc:\n",
        "            feature = gfcc_(signal,sr,cep)\n",
        "\n",
        "        quotient = int(feature.shape[0]/len_audio)\n",
        "        end_index = feature.shape[0]-(feature.shape[0]%len_audio)\n",
        "        r = feature[0:end_index,:]\n",
        "        re_data = np.reshape(r,(-1,len_audio,ceps)) #b = np.reshape(a, (8, 3, -1))\n",
        "        data=np.concatenate((data,re_data),axis=0)\n",
        "        \n",
        "        re_label = label(user,quotient)\n",
        "        data_label = np.concatenate((data_label,re_label),axis=0)\n",
        "    return data[1:,:,:],data_label"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "atS_es2Ame3Z",
        "colab_type": "code",
        "outputId": "07be0768-df19-4995-c1d9-8e1f84b69966",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 51
        }
      },
      "source": [
        "%%time\n",
        "def write_(x,y):\n",
        "    with open('mfcc_data.csv', 'a') as mfcc_dataFile:\n",
        "        writer = csv.writer(mfcc_dataFile)\n",
        "        writer.writerows(x)\n",
        "    mfcc_dataFile.close()\n",
        "    with open('mfcc_label.csv', 'a') as mfcc_labelFile:\n",
        "        writer = csv.writer(mfcc_labelFile)\n",
        "        writer.writerows(y)\n",
        "    mfcc_labelFile.close()"
      ],
      "execution_count": 16,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "CPU times: user 5 µs, sys: 0 ns, total: 5 µs\n",
            "Wall time: 10 µs\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "dAmw3NFOEp8p",
        "colab_type": "code",
        "outputId": "fd629a6a-339d-4854-c3b5-6775459ba80d",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 102
        }
      },
      "source": [
        "%%time\n",
        "x,y = s2c(mfcc=False,gmfcc=True,gfcc=False,cep=20)\n",
        "\n",
        "#write_(x,y)\n",
        "print(x.shape,y.shape)\n",
        "\n",
        "\n",
        "u = np.unique(y)\n",
        "num_classes = len(u)\n",
        "#print(u)\n",
        "d = dict()\n",
        "for enu,key in enumerate(u,0):\n",
        "    d[key]=enu\n",
        "print(d)\n",
        "for i in range(len(y)):\n",
        "    y[i] = d[y[i]]"
      ],
      "execution_count": 17,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "661it [00:59, 11.06it/s]"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "(2950, 148, 40) (2950,)\n",
            "{'1272': 0, '1462': 1, '1673': 2, '174': 3, '1919': 4, '251': 5, '422': 6, '652': 7, '777': 8, '84': 9}\n",
            "CPU times: user 1min 3s, sys: 46.7 s, total: 1min 50s\n",
            "Wall time: 59.8 s\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "gbqqVr_6Ep9d",
        "colab_type": "code",
        "outputId": "e177df22-4759-49e1-9657-7994fee14246",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 187
        }
      },
      "source": [
        "X  = x.reshape(x.shape[0],x.shape[1],x.shape[2],1)\n",
        "print(X.shape)\n",
        "import keras\n",
        "\n",
        "y_binary = keras.utils.to_categorical(y, num_classes = 10)\n",
        "#y_binary = np.asarray(list(map(int,y)))\n",
        "print(y_binary)\n",
        "\n",
        "x_train, x_test, y_train, y_test = train_test_split(X,y_binary ,test_size=0.20, shuffle= True)\n",
        "\n",
        "print(\"Training shape: {0}, Training label shape : {1}\".format(x_train.shape,y_train.shape))\n",
        "print(\"Training shape: {0}, Training label shape : {1}\".format(x_test.shape,y_test.shape))"
      ],
      "execution_count": 18,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "(2950, 148, 40, 1)\n",
            "[[0. 0. 0. ... 0. 0. 0.]\n",
            " [0. 0. 0. ... 0. 0. 0.]\n",
            " [0. 0. 0. ... 0. 0. 0.]\n",
            " ...\n",
            " [0. 0. 1. ... 0. 0. 0.]\n",
            " [0. 0. 1. ... 0. 0. 0.]\n",
            " [0. 0. 1. ... 0. 0. 0.]]\n",
            "Training shape: (2360, 148, 40, 1), Training label shape : (2360, 10)\n",
            "Training shape: (590, 148, 40, 1), Training label shape : (590, 10)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "FF6XbgMm7MGZ",
        "colab_type": "text"
      },
      "source": [
        "# Plotting plot :"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "mMj-6gUL7UYB",
        "colab_type": "text"
      },
      "source": [
        "\n",
        "\n",
        "1.   **accuracy_plot()** : This function plot graph between train accuracy vs validation/test accuracy\n",
        "2.   **loss_plot():** This function plot graph between train loss vs validation loss\n",
        "\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "xpHMjtZGEp_9",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "import keras\n",
        "from matplotlib import pyplot as plt\n",
        "def accuracy_plot(history):\n",
        "    #plt.figure(figsize=(12,8))\n",
        "    plt.plot(history.history['accuracy'])\n",
        "    plt.plot(history.history['val_accuracy'])\n",
        "    plt.title('model accuracy')\n",
        "    plt.ylabel('accuracy')\n",
        "    plt.xlabel('epoch')\n",
        "    plt.legend(['train', 'val'], loc='upper left')\n",
        "    plt.grid()\n",
        "    plt.show()\n",
        "\n",
        "def loss_plot(history):\n",
        "    #plt.figure(figsize=(12,8))\n",
        "    plt.plot(history.history['loss'])\n",
        "    plt.plot(history.history['val_loss'])\n",
        "    plt.title('model loss')\n",
        "    plt.ylabel('loss')\n",
        "    plt.xlabel('epoch')\n",
        "    plt.legend(['train', 'val'], loc='upper left')\n",
        "    plt.grid()\n",
        "    plt.show()"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "kzuVliqNEqBd",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "mNwKuBxHgY0d",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 54
        },
        "outputId": "38c1888c-349e-4d3e-a864-1351bfd09772"
      },
      "source": [
        "\"\"\"\n",
        "def audioLengthInSecond(sr,m,t_s):#t_s will gives time in second[1 for 1sec, 2 for 2sec]\n",
        "    t = round(sr/512) # Number of frames in one second\n",
        "    totalFrame = t*t_s # Number of frames in one second * time(second)\n",
        "    t1 = round(m.shape[0]/totalFrame) # total_audio_time/ t_s\n",
        "    #print(time_in_sec)\n",
        "    #print(\" Actual audio length in second :\",t1,\"\\n\",\"Number of frames in one second :\",t)\n",
        "    j=0\n",
        "    total = totalFrame*m.shape[1]\n",
        "    result = np.zeros([1,total])\n",
        "    #print(result.shape)\n",
        "    for i in range(t1):\n",
        "            temp = m[j:j+totalFrame,:].reshape(1,-1)\n",
        "            diff = total-temp.shape[1]\n",
        "            if round(total*0.3)>diff: # if more 30% padding need done then we will not add that audio file\n",
        "                #print(diff)\n",
        "                if diff!=0:\n",
        "                    temp = np.concatenate((temp[0,:],temp[0,diff:diff+diff])).reshape(1,-1)\n",
        "                #print(\"temp\",temp.shape)\n",
        "                result = np.concatenate((result,temp))\n",
        "                j += totalFrame\n",
        "            else:\n",
        "                pass\n",
        "    return result[1:,:]\n",
        "t_s=2\n",
        "r = audioLengthInSecond(sr,m,t_s)\n",
        "print(r.shape)\n",
        "\"\"\""
      ],
      "execution_count": 20,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "'\\ndef audioLengthInSecond(sr,m,t_s):#t_s will gives time in second[1 for 1sec, 2 for 2sec]\\n    t = round(sr/512) # Number of frames in one second\\n    totalFrame = t*t_s # Number of frames in one second * time(second)\\n    t1 = round(m.shape[0]/totalFrame) # total_audio_time/ t_s\\n    #print(time_in_sec)\\n    #print(\" Actual audio length in second :\",t1,\"\\n\",\"Number of frames in one second :\",t)\\n    j=0\\n    total = totalFrame*m.shape[1]\\n    result = np.zeros([1,total])\\n    #print(result.shape)\\n    for i in range(t1):\\n            temp = m[j:j+totalFrame,:].reshape(1,-1)\\n            diff = total-temp.shape[1]\\n            if round(total*0.3)>diff: # if more 30% padding need done then we will not add that audio file\\n                #print(diff)\\n                if diff!=0:\\n                    temp = np.concatenate((temp[0,:],temp[0,diff:diff+diff])).reshape(1,-1)\\n                #print(\"temp\",temp.shape)\\n                result = np.concatenate((result,temp))\\n                j += totalFrame\\n            else:\\n                pass\\n    return result[1:,:]\\nt_s=2\\nr = audioLengthInSecond(sr,m,t_s)\\nprint(r.shape)\\n'"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 20
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "d1HwaUks6gKY",
        "colab_type": "text"
      },
      "source": [
        "# Creating CNN Architecture :"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "cqpW8dQTnKt-",
        "colab_type": "code",
        "outputId": "5bbfcca5-c243-4d31-e93c-26895b77e4a8",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "source": [
        "%tensorflow_version 2.x\n",
        "import tensorflow \n",
        "device_name = tensorflow.test.gpu_device_name()\n",
        "if device_name != '/device:GPU:0':\n",
        "  raise SystemError('GPU device not found')\n",
        "print('Found GPU at: {}'.format(device_name))"
      ],
      "execution_count": 21,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Found GPU at: /device:GPU:0\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "zQfNxWMM1L5F",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 54
        },
        "outputId": "a83f1dfa-fe2b-44f1-d48f-989b34b7a80b"
      },
      "source": [
        "from tensorflow.keras.optimizers import SGD\n",
        "from tensorflow.python.keras.callbacks import ModelCheckpoint\n",
        "\"\"\"\n",
        "def CNN_model_training(x_train,y_train,x_text,y_test,output_size,epoch = 500,batch_sizes = 16):\n",
        "\n",
        "  model = Sequential()\n",
        "  model.add(Conv2D(32, (3, 3), activation='relu', input_shape=(x.shape[1],x.shape[2],1)))\n",
        "\n",
        "  model.add(Conv2D(32, (3, 3), activation='relu'))\n",
        "  model.add(MaxPooling2D(pool_size=(2, 2)))\n",
        "  model.add(Dropout(0.5))\n",
        "\n",
        "  model.add(Conv2D(64, (3, 3), activation='relu'))\n",
        "\n",
        "  model.add(Conv2D(64, (3, 3), activation='relu'))\n",
        "  model.add(MaxPooling2D(pool_size=(2, 2)))\n",
        "  model.add(Dropout(0.5))\n",
        "  model.add(Flatten())\n",
        "\n",
        "  model.add(Dense(256, activation='relu'))\n",
        "  model.add(Dropout(0.5))\n",
        "\n",
        "  model.add(Dense(output_size, activation='softmax'))\n",
        "  opt = SGD(lr=0.001, momentum=0.9)                        #optimizer=keras.optimizers.Adam()\n",
        "  model.compile(loss=keras.losses.categorical_crossentropy, optimizer=opt, metrics=['accuracy'])\n",
        "  print(model.summary())\n",
        "  \n",
        "  # Saving the model that performed the best on the validation set\n",
        "  checkpoint = ModelCheckpoint(filepath='Model.v_2.hdf5', save_best_only=True, verbose=0)\n",
        "\n",
        "\n",
        "\n",
        "  history = model.fit(x_train, y_train, batch_size = batch_sizes, epochs=epoch, \n",
        "                      validation_data=(x_test, y_test), verbose=0, callbacks=[checkpoint])\n",
        "  accuracy_plot(history)\n",
        "  loss_plot(history)                    \n",
        "\"\"\""
      ],
      "execution_count": 22,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "\"\\ndef CNN_model_training(x_train,y_train,x_text,y_test,output_size,epoch = 500,batch_sizes = 16):\\n\\n  model = Sequential()\\n  model.add(Conv2D(32, (3, 3), activation='relu', input_shape=(x.shape[1],x.shape[2],1)))\\n\\n  model.add(Conv2D(32, (3, 3), activation='relu'))\\n  model.add(MaxPooling2D(pool_size=(2, 2)))\\n  model.add(Dropout(0.5))\\n\\n  model.add(Conv2D(64, (3, 3), activation='relu'))\\n\\n  model.add(Conv2D(64, (3, 3), activation='relu'))\\n  model.add(MaxPooling2D(pool_size=(2, 2)))\\n  model.add(Dropout(0.5))\\n  model.add(Flatten())\\n\\n  model.add(Dense(256, activation='relu'))\\n  model.add(Dropout(0.5))\\n\\n  model.add(Dense(output_size, activation='softmax'))\\n  opt = SGD(lr=0.001, momentum=0.9)                        #optimizer=keras.optimizers.Adam()\\n  model.compile(loss=keras.losses.categorical_crossentropy, optimizer=opt, metrics=['accuracy'])\\n  print(model.summary())\\n  \\n  # Saving the model that performed the best on the validation set\\n  checkpoint = ModelCheckpoint(filepath='Model.v_2.hdf5', save_best_only=True, verbose=0)\\n\\n\\n\\n  history = model.fit(x_train, y_train, batch_size = batch_sizes, epochs=epoch, \\n                      validation_data=(x_test, y_test), verbose=0, callbacks=[checkpoint])\\n  accuracy_plot(history)\\n  loss_plot(history)                    \\n\""
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 22
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "EulSDA-nS7lu",
        "colab_type": "code",
        "outputId": "2f3e8e44-a09c-4e00-b9d3-fa166644f314",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 578
        }
      },
      "source": [
        "from tensorflow.keras.optimizers import SGD\n",
        "\n",
        "model = Sequential()\n",
        "model.add(Conv2D(32, (3, 3), activation='relu', input_shape=(x.shape[1],x.shape[2],1)))\n",
        "\n",
        "model.add(Conv2D(32, (3, 3), activation='relu'))\n",
        "model.add(MaxPooling2D(pool_size=(2, 2)))\n",
        "model.add(Dropout(0.5))\n",
        "\n",
        "model.add(Conv2D(64, (3, 3), activation='relu'))\n",
        "\n",
        "model.add(Conv2D(64, (3, 3), activation='relu'))\n",
        "model.add(MaxPooling2D(pool_size=(2, 2)))\n",
        "model.add(Dropout(0.5))\n",
        "model.add(Flatten())\n",
        "\n",
        "model.add(Dense(256, activation='relu'))\n",
        "model.add(Dropout(0.5))\n",
        "\n",
        "model.add(Dense(10, activation='softmax'))\n",
        "opt = SGD(lr=0.001, momentum=0.9)                        #optimizer=keras.optimizers.Adam()\n",
        "model.compile(loss=keras.losses.categorical_crossentropy, optimizer=opt, metrics=['accuracy'])\n",
        "print(model.summary())\n",
        "\n",
        "\n"
      ],
      "execution_count": 23,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Model: \"sequential\"\n",
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "conv2d (Conv2D)              (None, 146, 38, 32)       320       \n",
            "_________________________________________________________________\n",
            "conv2d_1 (Conv2D)            (None, 144, 36, 32)       9248      \n",
            "_________________________________________________________________\n",
            "max_pooling2d (MaxPooling2D) (None, 72, 18, 32)        0         \n",
            "_________________________________________________________________\n",
            "dropout (Dropout)            (None, 72, 18, 32)        0         \n",
            "_________________________________________________________________\n",
            "conv2d_2 (Conv2D)            (None, 70, 16, 64)        18496     \n",
            "_________________________________________________________________\n",
            "conv2d_3 (Conv2D)            (None, 68, 14, 64)        36928     \n",
            "_________________________________________________________________\n",
            "max_pooling2d_1 (MaxPooling2 (None, 34, 7, 64)         0         \n",
            "_________________________________________________________________\n",
            "dropout_1 (Dropout)          (None, 34, 7, 64)         0         \n",
            "_________________________________________________________________\n",
            "flatten (Flatten)            (None, 15232)             0         \n",
            "_________________________________________________________________\n",
            "dense (Dense)                (None, 256)               3899648   \n",
            "_________________________________________________________________\n",
            "dropout_2 (Dropout)          (None, 256)               0         \n",
            "_________________________________________________________________\n",
            "dense_1 (Dense)              (None, 10)                2570      \n",
            "=================================================================\n",
            "Total params: 3,967,210\n",
            "Trainable params: 3,967,210\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n",
            "None\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "0C4e0DA7Ep_K",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "%%time\n",
        "epoch = 100\n",
        "batch_sizes = 16\n",
        "from tensorflow.python.keras.callbacks import ModelCheckpoint\n",
        "# Saving the model that performed the best on the validation set\n",
        "checkpoint = ModelCheckpoint(filepath='Model.v_2.hdf5', save_best_only=True, verbose=0)\n",
        "\n",
        "\n",
        "\n",
        "history = model.fit(x_train, y_train, batch_size = batch_sizes, epochs=epoch, \n",
        "                    validation_data=(x_test, y_test), verbose=1, callbacks=[checkpoint])"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "RwEThfog4WjM",
        "colab_type": "code",
        "outputId": "a7422731-5f3d-4013-b3a5-83162232e1d1",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 51
        }
      },
      "source": [
        "#Loading the model that performed the best on the validation set\n",
        "model.load_weights('Model.v_2.hdf5')\n",
        "\n",
        "# Testing the model on the Test data\n",
        "(loss, accuracy) = model.evaluate(x_test, y_test, batch_size=16, verbose=1)\n",
        "\n",
        "print('Accuracy on test data: {:.2f}%'.format(accuracy * 100))"
      ],
      "execution_count": 25,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "37/37 [==============================] - 0s 7ms/step - loss: 0.1673 - accuracy: 0.9525\n",
            "Accuracy on test data: 95.25%\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "UGOIOWkoZBRf",
        "colab_type": "code",
        "outputId": "67cf6dad-f399-47cc-ccb7-7a8ed8838fc1",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 573
        }
      },
      "source": [
        "accuracy_plot(history)\n",
        "loss_plot(history)"
      ],
      "execution_count": 26,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "VguE_JJZZqoG",
        "colab_type": "code",
        "cellView": "form",
        "colab": {}
      },
      "source": [
        "#@title\n",
        "from sklearn.metrics import accuracy_score\n",
        "from sklearn.metrics import precision_score\n",
        "from sklearn.metrics import recall_score\n",
        "from sklearn.metrics import f1_score\n",
        "from sklearn.metrics import cohen_kappa_score\n",
        "from sklearn.metrics import roc_auc_score\n",
        "from sklearn.metrics import confusion_matrix"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "QUMQL7i0ZGnF",
        "colab_type": "code",
        "cellView": "form",
        "colab": {}
      },
      "source": [
        "#@title\n",
        "yhat_probs = model.predict(x_test, verbose=0)  #np.argmax(model.predict(x), axis=-1)\n",
        "# predict crisp classes for test set\n",
        "yhat_classes = model.predict_classes(x_test, verbose=0)\n",
        "print(yhat_probs.shape,yhat_classes.shape)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "siPeFSBIZPc_",
        "colab_type": "code",
        "cellView": "form",
        "colab": {}
      },
      "source": [
        "#@title\n",
        "# reduce to 1d array\n",
        "yhat_probs = yhat_probs[:, 0]\n",
        "#yhat_classes = yhat_classes[:, 0]"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "kh65OShvZcuS",
        "colab_type": "code",
        "cellView": "form",
        "colab": {}
      },
      "source": [
        "#@title\n",
        "# accuracy: (tp + tn) / (p + n)\n",
        "accuracy = accuracy_score(y_test, yhat_classes)\n",
        "print('Accuracy: %f' % accuracy)\n",
        "# precision tp / (tp + fp)\n",
        "precision = precision_score(y_test, yhat_classes)\n",
        "print('Precision: %f' % precision)\n",
        "# recall: tp / (tp + fn)\n",
        "recall = recall_score(y_test, yhat_classes)\n",
        "print('Recall: %f' % recall)\n",
        "# f1: 2 tp / (2 tp + fp + fn)\n",
        "f1 = f1_score(y_test, yhat_classes)\n",
        "print('F1 score: %f' % f1)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "sM5fF4DXZi3M",
        "colab_type": "code",
        "cellView": "form",
        "colab": {}
      },
      "source": [
        "#@title\n",
        "# kappa\n",
        "kappa = cohen_kappa_score(y_test, yhat_classes)\n",
        "print('Cohens kappa: %f' % kappa)\n",
        "# ROC AUC\n",
        "auc = roc_auc_score(y_test, yhat_probs)\n",
        "print('ROC AUC: %f' % auc)\n",
        "# confusion matrix\n",
        "matrix = confusion_matrix(y_test, yhat_classes)\n",
        "print(matrix)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "gdSAF2cy3E0X",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "import os, signal\n",
        "os.kill(os.getpid(), signal.SIGKILL)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "qlUm3WDCPrt_",
        "colab_type": "text"
      },
      "source": [
        "# VGG16"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "BiyIHBtrPu2B",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "\"\"\"model1 = Sequential()\n",
        "model1.add(Conv2D(input_shape=(x.shape[1],x.shape[2],1),filters=64,kernel_size=(3,3),padding=\"same\", activation=\"relu\"))\n",
        "model1.add(Conv2D(filters=64,kernel_size=(3,3),padding=\"same\", activation=\"relu\"))\n",
        "\n",
        "model1.add(MaxPooling2D(pool_size=(2,2),strides=(2,2)))\n",
        "\n",
        "model1.add(Conv2D(filters=128, kernel_size=(3,3), padding=\"same\", activation=\"relu\"))\n",
        "model1.add(Conv2D(filters=128, kernel_size=(3,3), padding=\"same\", activation=\"relu\"))\n",
        "\n",
        "model1.add(MaxPooling2D(pool_size=(2,2),strides=(1,1)))\n",
        "\n",
        "model1.add(Conv2D(filters=256, kernel_size=(3,3), padding=\"same\", activation=\"relu\"))\n",
        "model1.add(Conv2D(filters=256, kernel_size=(3,3), padding=\"same\", activation=\"relu\"))\n",
        "model1.add(Conv2D(filters=256, kernel_size=(3,3), padding=\"same\", activation=\"relu\"))\n",
        "\n",
        "model1.add(MaxPooling2D(pool_size=(2,2),strides=(2,2)))\n",
        "\n",
        "model1.add(Conv2D(filters=512, kernel_size=(3,3), pāadding=\"same\", activation=\"relu\"))\n",
        "model1.add(Conv2D(filters=512, kernel_size=(3,3), padding=\"same\", activation=\"relu\"))\n",
        "model1.add(Conv2D(filters=512, kernel_size=(3,3), padding=\"same\", activation=\"relu\"))\n",
        "\n",
        "model1.add(MaxPooling2D(pool_size=(2,2),strides=(2,2)))\n",
        "\n",
        "model1.add(Conv2D(filters=128, kernel_size=(3,3), padding=\"same\", activation=\"relu\"))\n",
        "model1.add(Conv2D(filters=128, kernel_size=(3,3), padding=\"same\", activation=\"relu\"))\n",
        "model1.add(Conv2D(filters=512, kernel_size=(3,3), padding=\"same\", activation=\"relu\"))\n",
        "\n",
        "model1.add(MaxPooling2D(pool_size=(2,2),strides=(2,2)))\n",
        "\n",
        "model1.add(Flatten())\n",
        "model1.add(Dense(units=128,activation=\"relu\"))\n",
        "model1.add(Dense(units=128,activation=\"relu\"))\n",
        "model1.add(Dense(units=40, activation=\"softmax\"))\n",
        "print(model1.summary())\n",
        "model1.compile(loss=\"sparse_categorical_crossentropy\", optimizer=\"adam\", metrics=['accuracy'])\n",
        "\"\"\""
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "PQXJf-duQKpE",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "\"\"\"%%time\n",
        "from keras.callbacks import ModelCheckpoint\n",
        "# Saving the model that performed the best on the validation set\n",
        "checkpoint = ModelCheckpoint(filepath='Model.v_2.hdf5', save_best_only=True, verbose=0)\n",
        "\n",
        "# Training the model for 40 epochs\n",
        "history = model1.fit(x_train, y_train, batch_size=32, epochs=200, \n",
        "                    validation_data=(x_test, y_test), verbose=1, callbacks=[checkpoint])\"\"\""
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Mwac_IauQMVC",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "\"\"\"# Loading the model that performed the best on the validation set\n",
        "model1.load_weights('Model.v_2.hdf5')\n",
        "\n",
        "# Testing the model on the Test data\n",
        "(loss, accuracy) = model1.evaluate(x_test, y_test, batch_size=128, verbose=1)\n",
        "\n",
        "print('Accuracy on test data: {:.2f}%'.format(accuracy * 100))\"\"\""
      ],
      "execution_count": 0,
      "outputs": []
    }
  ]
}