feature_extraction.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Gun Detection System for Urban Settings (_Feature extraction_)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "This notebook details the feature extraction techniques applied on the processed sound files. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import librosa\n",
    "import numpy as np\n",
    "from numpy import nan\n",
    "import pandas as pd\n",
    "import matplotlib\n",
    "from matplotlib import pyplot as plt\n",
    "import seaborn as sns\n",
    "import math\n",
    "from PIL import Image\n",
    "import os\n",
    "\n",
    "import warnings\n",
    "warnings.filterwarnings('ignore')\n",
    "matplotlib.use(\"Agg\")\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "First, some constant values are set to be used across the entire notebook"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "IMG_DIR = \"./mel-specs\"\n",
    "VAL_DIR = \"./mel-specs-val\"\n",
    "TEST_DIR = \"./mel-specs-test\"\n",
    "\n",
    "SAMPLE_RATE = 22050 # Justification for this value is provided later"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### II.IV Feature Extraction"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The splitted sets are first loaded into numpy array to prepare for extraction."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**_Loading data sets_**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "def load_x_and_y_helper(path):\n",
    "    x = []\n",
    "    y = []\n",
    "    \n",
    "    files = librosa.util.find_files(path, ext=['wav'])\n",
    "    files = np.asarray(files)\n",
    "    for file in files:\n",
    "        if '.wav' in file:\n",
    "            name = file.split('/').pop()\n",
    "            label = name.split('.')[0].split('-')[1];\n",
    "            wave_arr, sr = librosa.load(file, sr = SAMPLE_RATE, mono = True)\n",
    "            x = x + [wave_arr]\n",
    "            y = y + [label]\n",
    "    print('done!')\n",
    "    return np.array(x), np.array(y, dtype=int)\n",
    "\n",
    "\n",
    "def load_x_and_ys():\n",
    "    print('collecting training set...', end = \"\")\n",
    "    X_train, y_train = load_x_and_y_helper('./processed_data/train')\n",
    "    print('collecting validation set...', end = \"\")\n",
    "    X_val, y_val = load_x_and_y_helper('./processed_data/val')\n",
    "    print('collecting test set...', end = \"\")\n",
    "    X_test, y_test = load_x_and_y_helper('./processed_data/test')\n",
    "    \n",
    "    return X_train, y_train, X_val, y_val, X_test, y_test"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "collecting training set...done!\n",
      "collecting validation set...done!\n",
      "collecting test set...done!\n"
     ]
    }
   ],
   "source": [
    "X_train, y_train, X_val, y_val, X_test, y_test = load_x_and_ys()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "*PS: the y vactor now contains the ID of classes \\[0-9\\] instead of the categorical names*"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((7000, 44100), (7000,))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "X_train.shape, y_train.shape"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "For this project, two approaches will be implemented for comparison, **Mel-spectrogram classification** and **multi-feature stacking**.<br/>\n",
    "First, all audio arrays will be converted to Mel-spectrograms in preparation for the first approach. <br/>\n",
    "Then, relevant 1D feature vectors will be extracted from the same original sets and stacked in preparation for the second approach."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**_Mel-spectrogram images_**"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Using librosa, all audio slices will be converted into 2D mel-spectrogram matrix. The matrix will then be saved in the form of a heatmap-like image. The construction of those images takes a very long time so the resulted images will be saved to disk to be loaded later when the training phase starts. This will allow for more experimentation in the training phase without the need to recompute the images each time."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "When the resulting mel-spectrogram is images is saved, it is saved as a diagram with x-axis and y-axis, as well as an extra empty white border frame (see example below). All audios have the same length and sample rate so they all share the same scale of the x-axis and y-axis. Thus, the axes, labels, and border are redundant and may cause confusion for the CNN model, leading it to learn irrelevant features and lower the training quality. For these reasons, after saving the spectrogram images, the `resize_images` function is called to loop through the output images and crop the unecessary data, leaving only the spectrogram heatmap in the picture.  "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "def resize_images(base_path=IMG_DIR):\n",
    "    subfolders = [f.path for f in os.scandir(base_path) if f.is_dir()]\n",
    "    images = []\n",
    "    \n",
    "    for fold_path in subfolders:\n",
    "        for file in os.listdir(fold_path):\n",
    "            if '.png' in file:\n",
    "                path = \"{}/{}\".format(fold_path, file)\n",
    "                im = Image.open(path)\n",
    "                width, height = im.size\n",
    "\n",
    "                left = 81\n",
    "                top = 59\n",
    "                right = width - 64\n",
    "                bottom = height - 53\n",
    "\n",
    "                im1 = im.crop((left, top, right, bottom))\n",
    "                im1.save(path)\n",
    "                images += [im1]\n",
    "                \n",
    "                im.close()\n",
    "    return images\n",
    "\n",
    "def convert_to_mel_specs(X, y, base_path=IMG_DIR, show_example=False):\n",
    "    if not os.path.exists(base_path):\n",
    "        os.mkdir(base_path)\n",
    "        \n",
    "    for label in label_map.keys():\n",
    "        sub_fold = \"{}/{}\".format(base_path, label)\n",
    "        if not os.path.exists(sub_fold):\n",
    "            os.mkdir(sub_fold)\n",
    "\n",
    "    X_mel_specs = []\n",
    "    for idx, wav in enumerate(X):\n",
    "        spec = librosa.feature.melspectrogram(y=wav.astype(float), sr=SAMPLE_RATE)\n",
    "        X_mel_specs = X_mel_specs + [spec]\n",
    "        \n",
    "        fig = plt.Figure()\n",
    "        ax = fig.add_subplot(111)\n",
    "        p = librosa.display.specshow(librosa.amplitude_to_db(spec, ref=np.max), ax=ax, y_axis='log', x_axis='time')\n",
    "        fig.savefig(\"{}/{}/{}-{}.png\".format(base_path, get_class_name(int(y[idx])), idx, y[idx]))\n",
    "\n",
    "    ims = resize_images(base_path)\n",
    "    \n",
    "    return X_mel_specs, ims"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "X_specs_train, images = convert_to_mel_specs(X_train, y_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Original photo of the mel-spectrogram:\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The final cropped image:\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<PIL.Image.Image image mode=RGBA size=495x368>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "print('Original photo of the mel-spectrogram:')\n",
    "p = librosa.display.specshow(X_specs_train[0], sr=SAMPLE_RATE, x_axis='time', y_axis='log');\n",
    "plt.show()\n",
    "print('The final cropped image:')\n",
    "images[0]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "*PS: Those two images are not of the same sound wave as PIL Image shuffles the order of the original set of images.*"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "X_specs_val = convert_to_mel_specs(X_val, y_val, VAL_DIR)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "X_specs_test = convert_to_mel_specs(X_test, y_test, TEST_DIR)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(7000, 128, 87)"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "org_shape = np.array(X_specs_train).shape\n",
    "org_shape"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The raw specs array is saved in csv file for future use"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "def save_specs_arr(arr, name):\n",
    "    np.savetxt(\"./processed_data/{}.txt\".format(name), np.array(arr).reshape(len(arr), -1))\n",
    "    \n",
    "def load_specs_arr(name):\n",
    "    arr = np.loadtxt(\"./processed_data/{}.txt\".format(name))\n",
    "    arr = arr.reshape(arr.shape[0], arr.shape[1] // org_shape[2], org_shape[2])\n",
    "    return arr"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "save_specs_arr(X_specs_train, 'training_mel_specs')\n",
    "save_specs_arr(X_specs_val, 'val_mel_specs')\n",
    "save_specs_arr(X_specs_test, 'test_mel_specs')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**_1D stacked vectors_**"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "For the second approach, a 1D vector is extracted from each audio clip representing this audio. This vector is the result of stacking different features describing the same audio wave. Five features will be computed from each audio: Mel-spectrogram, MFCCs, spectral contrast feature, chromagram and tonnetz. Each of these features will be then converted from a 2D matrix to a 1D array by taking the mean across the vertical axis. According to a study in [[1]](https://www.researchgate.net/publication/365912955_Multi-feature_stacking_order_impact_on_speech_emotion_recognition_performance), the orders in which these features are stacked is significantly impactful on the final result. Therefore, following their recommended order, the features are horizontally stacked in the following order: spectral contrast, tonnetz, chromagram, Mel-spectrogram, and MFCC. The resulting vectors are then inputted into the classifier."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "def extract_stacked_feats(X):\n",
    "    mfccs = librosa.feature.mfcc(y=X, sr=SAMPLE_RATE, n_mfcc=40)\n",
    "    _1D_mfcc = np.mean(mfccs.T, axis=0)\n",
    "    \n",
    "    chromagram = librosa.feature.chroma_stft(y=X, sr=SAMPLE_RATE)\n",
    "    _1D_chroma = np.mean(chromagram.T, axis=0)\n",
    "\n",
    "    mel_spec = librosa.feature.melspectrogram(y=X, sr=SAMPLE_RATE)\n",
    "    _1D_mel = np.mean(mel_spec.T, axis=0)\n",
    "\n",
    "    contrast = librosa.feature.spectral_contrast(y=X, sr=SAMPLE_RATE)\n",
    "    _1D_contrast = np.mean(contrast.T, axis=0)\n",
    "\n",
    "    harmonic = librosa.effects.harmonic(X)\n",
    "    tonnetz = librosa.feature.tonnetz(y=harmonic, sr=SAMPLE_RATE)\n",
    "    _1D_tonnetz = np.mean(tonnetz.T,axis=0)\n",
    "    \n",
    "    \n",
    "    return np.hstack([_1D_contrast, _1D_tonnetz, _1D_chroma, _1D_mel, _1D_mfcc])\n",
    "\n",
    "def extract_stacked_feature_arrays(X):\n",
    "    data = [];\n",
    "    \n",
    "    print('extracting', end=\"\")\n",
    "    progress = 0;\n",
    "    for idx in range(len(X)):\n",
    "        if(idx == int(progress * len(X))):\n",
    "            print('.', end=\"\")\n",
    "            progress += 0.1\n",
    "            \n",
    "        stack = extract_stacked_feats(X[idx])\n",
    "        data = data + [stack]\n",
    "    print('done!')\n",
    "    \n",
    "    return np.array(data)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training set: extracting...........done!\n",
      "Validation set: extracting...........done!\n",
      "Test set: extracting...........done!\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "(7000, 193)"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "print('Training set: ', end='')\n",
    "X_stack_train = extract_stacked_feature_arrays(X_train)\n",
    "print('Validation set: ', end='')\n",
    "X_stack_val = extract_stacked_feature_arrays(X_val)\n",
    "print('Test set: ', end='')\n",
    "X_stack_test = extract_stacked_feature_arrays(X_test)\n",
    "\n",
    "X_stack_train.shape"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The 1D feature vectors are saved in csv format for future use."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "def export_stacked_feature_arrays(X, y, split='train'):\n",
    "    path = './processed_data/stacked_features'\n",
    "    if not os.path.exists(path):\n",
    "        os.mkdir(path)\n",
    "    \n",
    "    print('exporting {} set...'.format(split), end=' ')\n",
    "    np.savetxt(\"{}/X_{}.csv\".format(path, split), X, delimiter=\",\")\n",
    "    y.tofile('{}/y_{}.csv'.format(path, split), sep = ',')\n",
    "    print('done!')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "exporting train set... done!\n",
      "exporting val set... done!\n",
      "exporting test set... done!\n"
     ]
    }
   ],
   "source": [
    "export_stacked_feature_arrays(X_stack_train, y_train)\n",
    "export_stacked_feature_arrays(X_stack_val, y_val, split='val')\n",
    "export_stacked_feature_arrays(X_stack_test, y_test, split='test')"
   ]
  }
 ],
 "metadata": {
  "language_info": {
   "name": "python"
  },
  "orig_nbformat": 4
 },
 "nbformat": 4,
 "nbformat_minor": 2
}