mfcc-cnn-voice-recognition.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "9a84ffa2",
   "metadata": {},
   "outputs": [],
   "source": [
    "import json\n",
    "import os\n",
    "import math\n",
    "import librosa\n",
    "\n",
    "DATASET_PATH = \"/Users/mohammad/Desktop/master/audio\"\n",
    "JSON_PATH = \"data.json\"\n",
    "SAMPLE_RATE = 44100\n",
    "TRACK_DURATION = 5 # measured in seconds\n",
    "SAMPLES_PER_TRACK = SAMPLE_RATE * TRACK_DURATION\n",
    "\n",
    "def save_mfcc(dataset_path, json_path, num_mfcc=13, n_fft=2048, hop_length=512, num_segments=2):\n",
    "    \"\"\"Extracts MFCCs from dataset and saves them into a json file along witgh specific labels.\n",
    "        :param dataset_path (str): Path to dataset\n",
    "        :param json_path (str): Path to json file used to save MFCCs\n",
    "        :param num_mfcc (int): Number of coefficients to extract\n",
    "        :param n_fft (int): Interval we consider to apply FFT. Measured in # of samples\n",
    "        :param hop_length (int): Sliding window for FFT. Measured in # of samples\n",
    "        :param: num_segments (int): Number of segments we want to divide sample tracks into\n",
    "        :return:\n",
    "        \"\"\"\n",
    "\n",
    "    # dictionary to store mapping, labels, and MFCCs\n",
    "    data = {\n",
    "        \"mapping\": [],\n",
    "        \"labels\": [],\n",
    "        \"mfcc\": []\n",
    "    }\n",
    "\n",
    "    samples_per_segment = int(SAMPLES_PER_TRACK / num_segments)\n",
    "    num_mfcc_vectors_per_segment = math.ceil(samples_per_segment / hop_length)\n",
    "\n",
    "    # loop through all genre sub-folder\n",
    "    for i, (dirpath, dirnames, filenames) in enumerate(os.walk(dataset_path)):\n",
    "\n",
    "        # ensure we're processing a genre sub-folder level\n",
    "        if dirpath is not dataset_path:\n",
    "\n",
    "            # save genre label (i.e., sub-folder name) in the mapping\n",
    "            semantic_label = dirpath.split(\"/\")[-1]\n",
    "            data[\"mapping\"].append(semantic_label)\n",
    "\n",
    "            # process all audio files in genre sub-dir\n",
    "            for f in filenames:\n",
    "\n",
    "            # load audio file\n",
    "                file_path = os.path.join(dirpath, f)\n",
    "                signal, sample_rate = librosa.load(file_path, sr=SAMPLE_RATE)\n",
    "\n",
    "                # process all segments of audio file\n",
    "                for d in range(num_segments):\n",
    "\n",
    "                    # calculate start and finish sample for current segment\n",
    "                    start = samples_per_segment * d\n",
    "                    finish = start + samples_per_segment\n",
    " \n",
    "                    # extract mfcc\n",
    "                    mfcc = librosa.feature.mfcc(signal[start:finish], sample_rate, n_mfcc=num_mfcc, n_fft=n_fft, hop_length=hop_length)\n",
    "                    mfcc = mfcc.T\n",
    "\n",
    "                    # store only mfcc feature with expected number of vectors\n",
    "                    if len(mfcc) == num_mfcc_vectors_per_segment:\n",
    "                        data[\"mfcc\"].append(mfcc.tolist())\n",
    "                        data[\"labels\"].append(i-1)\n",
    "\n",
    "    # save MFCCs to json file\n",
    "    with open(json_path, \"w\") as fp:\n",
    "        json.dump(data, fp, indent=4)\n",
    "\n",
    "        \n",
    "if __name__ == \"__main__\":\n",
    "    save_mfcc(DATASET_PATH, JSON_PATH, num_segments=2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "b41d43e0",
   "metadata": {},
   "outputs": [],
   "source": [
    "import json\n",
    "import numpy as np\n",
    "from sklearn.model_selection import train_test_split\n",
    "import tensorflow.keras as keras\n",
    "\n",
    "# path to json file that stores MFCCs and genre labels for each processed segment\n",
    "DATA_PATH = \"data.json\"\n",
    "\n",
    "def load_data(data_path):\n",
    "    with open(data_path, \"r\") as fp:\n",
    "        data = json.load(fp)\n",
    "\n",
    "    # convert lists to numpy arrays\n",
    "    X = np.array(data[\"mfcc\"])\n",
    "    y = np.array(data[\"labels\"])\n",
    "\n",
    "    print(\"Data succesfully loaded!\")\n",
    "\n",
    "    return  X, y\n",
    "\n",
    "\n",
    "if __name__ == \"__main__\":\n",
    "\n",
    "    # load data\n",
    "    X, y = load_data(DATA_PATH)\n",
    "\n",
    "    # create train/test split\n",
    "    X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.1)\n",
    "\n",
    "    # build network topology\n",
    "    model = keras.Sequential([\n",
    "        # input layer\n",
    "        keras.layers.Flatten(input_shape=(X.shape[1], X.shape[2])),\n",
    "\n",
    "        # 1st dense layer\n",
    "        keras.layers.Dense(512, activation='relu'),\n",
    "\n",
    "        # 2nd dense layer\n",
    "        keras.layers.Dense(256, activation='relu'),\n",
    "\n",
    "        # 3rd dense layer\n",
    "        keras.layers.Dense(64, activation='relu'),\n",
    "\n",
    "        # output layer\n",
    "        keras.layers.Dense(11, activation='softmax')\n",
    "    ])\n",
    "\n",
    "    # compile model\n",
    "    optimiser = keras.optimizers.Adam(learning_rate=0.0001)\n",
    "    model.compile(optimizer=optimiser,\n",
    "                  loss='sparse_categorical_crossentropy',\n",
    "                  metrics=['accuracy'])\n",
    "\n",
    "    model.summary()\n",
    "\n",
    "    # train model\n",
    "    history = model.fit(X_train, y_train, validation_data=(X_test, y_test), batch_size=32, epochs=50)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "88013c9b",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Data succesfully loaded!\n",
      "Model: \"sequential_1\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "flatten_1 (Flatten)          (None, 2808)              0         \n",
      "_________________________________________________________________\n",
      "dense_4 (Dense)              (None, 512)               1438208   \n",
      "_________________________________________________________________\n",
      "dropout_3 (Dropout)          (None, 512)               0         \n",
      "_________________________________________________________________\n",
      "dense_5 (Dense)              (None, 256)               131328    \n",
      "_________________________________________________________________\n",
      "dropout_4 (Dropout)          (None, 256)               0         \n",
      "_________________________________________________________________\n",
      "dense_6 (Dense)              (None, 64)                16448     \n",
      "_________________________________________________________________\n",
      "dropout_5 (Dropout)          (None, 64)                0         \n",
      "_________________________________________________________________\n",
      "dense_7 (Dense)              (None, 3)                 195       \n",
      "=================================================================\n",
      "Total params: 1,586,179\n",
      "Trainable params: 1,586,179\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n",
      "Epoch 1/100\n",
      "6/6 [==============================] - 1s 54ms/step - loss: 70.0413 - accuracy: 0.3892 - val_loss: 22.5264 - val_accuracy: 0.6316\n",
      "Epoch 2/100\n",
      "6/6 [==============================] - 0s 20ms/step - loss: 57.3157 - accuracy: 0.4268 - val_loss: 20.9433 - val_accuracy: 0.4211\n",
      "Epoch 3/100\n",
      "6/6 [==============================] - 0s 23ms/step - loss: 48.2059 - accuracy: 0.4413 - val_loss: 13.8377 - val_accuracy: 0.5263\n",
      "Epoch 4/100\n",
      "6/6 [==============================] - 0s 22ms/step - loss: 30.9419 - accuracy: 0.4866 - val_loss: 13.4372 - val_accuracy: 0.5263\n",
      "Epoch 5/100\n",
      "6/6 [==============================] - 0s 21ms/step - loss: 33.3952 - accuracy: 0.4525 - val_loss: 10.3778 - val_accuracy: 0.5789\n",
      "Epoch 6/100\n",
      "6/6 [==============================] - 0s 20ms/step - loss: 27.8693 - accuracy: 0.5432 - val_loss: 9.7074 - val_accuracy: 0.5789\n",
      "Epoch 7/100\n",
      "6/6 [==============================] - 0s 20ms/step - loss: 26.7161 - accuracy: 0.5168 - val_loss: 5.8341 - val_accuracy: 0.4737\n",
      "Epoch 8/100\n",
      "6/6 [==============================] - 0s 23ms/step - loss: 29.3984 - accuracy: 0.4735 - val_loss: 4.1265 - val_accuracy: 0.5263\n",
      "Epoch 9/100\n",
      "6/6 [==============================] - 0s 22ms/step - loss: 27.4002 - accuracy: 0.5418 - val_loss: 3.8593 - val_accuracy: 0.5263\n",
      "Epoch 10/100\n",
      "6/6 [==============================] - 0s 17ms/step - loss: 23.9187 - accuracy: 0.4909 - val_loss: 3.9498 - val_accuracy: 0.5789\n",
      "Epoch 11/100\n",
      "6/6 [==============================] - 0s 23ms/step - loss: 21.2867 - accuracy: 0.5247 - val_loss: 4.1919 - val_accuracy: 0.5789\n",
      "Epoch 12/100\n",
      "6/6 [==============================] - 0s 19ms/step - loss: 17.3459 - accuracy: 0.5606 - val_loss: 4.7207 - val_accuracy: 0.5789\n",
      "Epoch 13/100\n",
      "6/6 [==============================] - 0s 21ms/step - loss: 20.0270 - accuracy: 0.5297 - val_loss: 4.1269 - val_accuracy: 0.6316\n",
      "Epoch 14/100\n",
      "6/6 [==============================] - 0s 24ms/step - loss: 17.4417 - accuracy: 0.5471 - val_loss: 3.4591 - val_accuracy: 0.7368\n",
      "Epoch 15/100\n",
      "6/6 [==============================] - 0s 19ms/step - loss: 17.0282 - accuracy: 0.5102 - val_loss: 3.8423 - val_accuracy: 0.7895\n",
      "Epoch 16/100\n",
      "6/6 [==============================] - 0s 20ms/step - loss: 18.2966 - accuracy: 0.5445 - val_loss: 4.8959 - val_accuracy: 0.6316\n",
      "Epoch 17/100\n",
      "6/6 [==============================] - 0s 22ms/step - loss: 16.9855 - accuracy: 0.5213 - val_loss: 5.9277 - val_accuracy: 0.5789\n",
      "Epoch 18/100\n",
      "6/6 [==============================] - 0s 18ms/step - loss: 14.5736 - accuracy: 0.6349 - val_loss: 4.5990 - val_accuracy: 0.6316\n",
      "Epoch 19/100\n",
      "6/6 [==============================] - 0s 15ms/step - loss: 12.3705 - accuracy: 0.5730 - val_loss: 3.6860 - val_accuracy: 0.7368\n",
      "Epoch 20/100\n",
      "6/6 [==============================] - 0s 14ms/step - loss: 15.8892 - accuracy: 0.5357 - val_loss: 3.2021 - val_accuracy: 0.6316\n",
      "Epoch 21/100\n",
      "6/6 [==============================] - 0s 17ms/step - loss: 11.9214 - accuracy: 0.6329 - val_loss: 2.9249 - val_accuracy: 0.6316\n",
      "Epoch 22/100\n",
      "6/6 [==============================] - 0s 15ms/step - loss: 11.0470 - accuracy: 0.6576 - val_loss: 2.6872 - val_accuracy: 0.5263\n",
      "Epoch 23/100\n",
      "6/6 [==============================] - 0s 17ms/step - loss: 14.2298 - accuracy: 0.6221 - val_loss: 2.7399 - val_accuracy: 0.5263\n",
      "Epoch 24/100\n",
      "6/6 [==============================] - 0s 16ms/step - loss: 14.5645 - accuracy: 0.5575 - val_loss: 3.2677 - val_accuracy: 0.5263\n",
      "Epoch 25/100\n",
      "6/6 [==============================] - 0s 20ms/step - loss: 12.5282 - accuracy: 0.5831 - val_loss: 3.5174 - val_accuracy: 0.5263\n",
      "Epoch 26/100\n",
      "6/6 [==============================] - 0s 20ms/step - loss: 12.2919 - accuracy: 0.5368 - val_loss: 3.4630 - val_accuracy: 0.5789\n",
      "Epoch 27/100\n",
      "6/6 [==============================] - 0s 20ms/step - loss: 10.1428 - accuracy: 0.5721 - val_loss: 3.8090 - val_accuracy: 0.5263\n",
      "Epoch 28/100\n",
      "6/6 [==============================] - 0s 18ms/step - loss: 11.9670 - accuracy: 0.5200 - val_loss: 3.5333 - val_accuracy: 0.4737\n",
      "Epoch 29/100\n",
      "6/6 [==============================] - 0s 17ms/step - loss: 10.1297 - accuracy: 0.5564 - val_loss: 3.1111 - val_accuracy: 0.5789\n",
      "Epoch 30/100\n",
      "6/6 [==============================] - 0s 16ms/step - loss: 10.7409 - accuracy: 0.5647 - val_loss: 2.1575 - val_accuracy: 0.6842\n",
      "Epoch 31/100\n",
      "6/6 [==============================] - 0s 21ms/step - loss: 10.0088 - accuracy: 0.6170 - val_loss: 1.9067 - val_accuracy: 0.7368\n",
      "Epoch 32/100\n",
      "6/6 [==============================] - 0s 16ms/step - loss: 9.3001 - accuracy: 0.5786 - val_loss: 2.4401 - val_accuracy: 0.6316\n",
      "Epoch 33/100\n",
      "6/6 [==============================] - 0s 14ms/step - loss: 12.2449 - accuracy: 0.5133 - val_loss: 2.6517 - val_accuracy: 0.5789\n",
      "Epoch 34/100\n",
      "6/6 [==============================] - 0s 14ms/step - loss: 8.3581 - accuracy: 0.6102 - val_loss: 3.2367 - val_accuracy: 0.6316\n",
      "Epoch 35/100\n",
      "6/6 [==============================] - 0s 14ms/step - loss: 8.7029 - accuracy: 0.5267 - val_loss: 3.2995 - val_accuracy: 0.5789\n",
      "Epoch 36/100\n",
      "6/6 [==============================] - 0s 15ms/step - loss: 7.7885 - accuracy: 0.5781 - val_loss: 3.4739 - val_accuracy: 0.6842\n",
      "Epoch 37/100\n",
      "6/6 [==============================] - 0s 14ms/step - loss: 6.9413 - accuracy: 0.6398 - val_loss: 3.3083 - val_accuracy: 0.5789\n",
      "Epoch 38/100\n",
      "6/6 [==============================] - 0s 15ms/step - loss: 6.9226 - accuracy: 0.6398 - val_loss: 2.8110 - val_accuracy: 0.6316\n",
      "Epoch 39/100\n",
      "6/6 [==============================] - 0s 14ms/step - loss: 7.5044 - accuracy: 0.5939 - val_loss: 2.9136 - val_accuracy: 0.6316\n",
      "Epoch 40/100\n",
      "6/6 [==============================] - 0s 15ms/step - loss: 7.7977 - accuracy: 0.6447 - val_loss: 3.2755 - val_accuracy: 0.6316\n",
      "Epoch 41/100\n",
      "6/6 [==============================] - 0s 23ms/step - loss: 8.9236 - accuracy: 0.6045 - val_loss: 4.1093 - val_accuracy: 0.6316\n",
      "Epoch 42/100\n",
      "6/6 [==============================] - 0s 21ms/step - loss: 8.1539 - accuracy: 0.6590 - val_loss: 3.7459 - val_accuracy: 0.5789\n",
      "Epoch 43/100\n",
      "6/6 [==============================] - 0s 19ms/step - loss: 8.1370 - accuracy: 0.6057 - val_loss: 3.5574 - val_accuracy: 0.6316\n",
      "Epoch 44/100\n",
      "6/6 [==============================] - 0s 17ms/step - loss: 7.7868 - accuracy: 0.5680 - val_loss: 2.8288 - val_accuracy: 0.6316\n",
      "Epoch 45/100\n",
      "6/6 [==============================] - 0s 19ms/step - loss: 7.5776 - accuracy: 0.5660 - val_loss: 2.3464 - val_accuracy: 0.6316\n",
      "Epoch 46/100\n",
      "6/6 [==============================] - 0s 17ms/step - loss: 7.4605 - accuracy: 0.6154 - val_loss: 2.5592 - val_accuracy: 0.7895\n",
      "Epoch 47/100\n",
      "6/6 [==============================] - 0s 14ms/step - loss: 6.4957 - accuracy: 0.5349 - val_loss: 2.5338 - val_accuracy: 0.7895\n",
      "Epoch 48/100\n",
      "6/6 [==============================] - 0s 17ms/step - loss: 6.1227 - accuracy: 0.6617 - val_loss: 2.6132 - val_accuracy: 0.7368\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 49/100\n",
      "6/6 [==============================] - 0s 15ms/step - loss: 5.9185 - accuracy: 0.6526 - val_loss: 2.3390 - val_accuracy: 0.6316\n",
      "Epoch 50/100\n",
      "6/6 [==============================] - 0s 20ms/step - loss: 5.9207 - accuracy: 0.7188 - val_loss: 2.1409 - val_accuracy: 0.6316\n",
      "Epoch 51/100\n",
      "6/6 [==============================] - 0s 19ms/step - loss: 5.8903 - accuracy: 0.6555 - val_loss: 2.1994 - val_accuracy: 0.5789\n",
      "Epoch 52/100\n",
      "6/6 [==============================] - 0s 26ms/step - loss: 6.9601 - accuracy: 0.6025 - val_loss: 2.1685 - val_accuracy: 0.5789\n",
      "Epoch 53/100\n",
      "6/6 [==============================] - 0s 26ms/step - loss: 5.6396 - accuracy: 0.6251 - val_loss: 2.1424 - val_accuracy: 0.6316\n",
      "Epoch 54/100\n",
      "6/6 [==============================] - 0s 20ms/step - loss: 6.3098 - accuracy: 0.6627 - val_loss: 1.9206 - val_accuracy: 0.5789\n",
      "Epoch 55/100\n",
      "6/6 [==============================] - 0s 25ms/step - loss: 6.2389 - accuracy: 0.6642 - val_loss: 2.2055 - val_accuracy: 0.5789\n",
      "Epoch 56/100\n",
      "6/6 [==============================] - 0s 25ms/step - loss: 5.0317 - accuracy: 0.6318 - val_loss: 2.2115 - val_accuracy: 0.7895\n",
      "Epoch 57/100\n",
      "6/6 [==============================] - 0s 22ms/step - loss: 5.4179 - accuracy: 0.6630 - val_loss: 2.2061 - val_accuracy: 0.7368\n",
      "Epoch 58/100\n",
      "6/6 [==============================] - 0s 22ms/step - loss: 5.3676 - accuracy: 0.6241 - val_loss: 2.1643 - val_accuracy: 0.7368\n",
      "Epoch 59/100\n",
      "6/6 [==============================] - 0s 22ms/step - loss: 4.5613 - accuracy: 0.6829 - val_loss: 2.0120 - val_accuracy: 0.7895\n",
      "Epoch 60/100\n",
      "6/6 [==============================] - 0s 22ms/step - loss: 5.2044 - accuracy: 0.5744 - val_loss: 1.6675 - val_accuracy: 0.7368\n",
      "Epoch 61/100\n",
      "6/6 [==============================] - 0s 24ms/step - loss: 3.9014 - accuracy: 0.6941 - val_loss: 1.8483 - val_accuracy: 0.7368\n",
      "Epoch 62/100\n",
      "6/6 [==============================] - 0s 25ms/step - loss: 4.7548 - accuracy: 0.6549 - val_loss: 1.8426 - val_accuracy: 0.7895\n",
      "Epoch 63/100\n",
      "6/6 [==============================] - 0s 21ms/step - loss: 4.4266 - accuracy: 0.6642 - val_loss: 1.7561 - val_accuracy: 0.7895\n",
      "Epoch 64/100\n",
      "6/6 [==============================] - 0s 27ms/step - loss: 6.2913 - accuracy: 0.6060 - val_loss: 1.5643 - val_accuracy: 0.8421\n",
      "Epoch 65/100\n",
      "6/6 [==============================] - 0s 22ms/step - loss: 4.6058 - accuracy: 0.6411 - val_loss: 1.6562 - val_accuracy: 0.7895\n",
      "Epoch 66/100\n",
      "6/6 [==============================] - 0s 20ms/step - loss: 5.2580 - accuracy: 0.6521 - val_loss: 1.6908 - val_accuracy: 0.7368\n",
      "Epoch 67/100\n",
      "6/6 [==============================] - 0s 24ms/step - loss: 3.3212 - accuracy: 0.7642 - val_loss: 1.7497 - val_accuracy: 0.7895\n",
      "Epoch 68/100\n",
      "6/6 [==============================] - 0s 20ms/step - loss: 4.5360 - accuracy: 0.6101 - val_loss: 1.7130 - val_accuracy: 0.6842\n",
      "Epoch 69/100\n",
      "6/6 [==============================] - 0s 20ms/step - loss: 3.8017 - accuracy: 0.6861 - val_loss: 1.7901 - val_accuracy: 0.7368\n",
      "Epoch 70/100\n",
      "6/6 [==============================] - 0s 16ms/step - loss: 3.7815 - accuracy: 0.6697 - val_loss: 1.9623 - val_accuracy: 0.6842\n",
      "Epoch 71/100\n",
      "6/6 [==============================] - 0s 15ms/step - loss: 4.3984 - accuracy: 0.6737 - val_loss: 1.9670 - val_accuracy: 0.7895\n",
      "Epoch 72/100\n",
      "6/6 [==============================] - 0s 16ms/step - loss: 4.6601 - accuracy: 0.6874 - val_loss: 1.9624 - val_accuracy: 0.7895\n",
      "Epoch 73/100\n",
      "6/6 [==============================] - 0s 15ms/step - loss: 5.0729 - accuracy: 0.5620 - val_loss: 2.0552 - val_accuracy: 0.7368\n",
      "Epoch 74/100\n",
      "6/6 [==============================] - 0s 14ms/step - loss: 3.6393 - accuracy: 0.7057 - val_loss: 2.3086 - val_accuracy: 0.6316\n",
      "Epoch 75/100\n",
      "6/6 [==============================] - 0s 20ms/step - loss: 4.3016 - accuracy: 0.7111 - val_loss: 2.1209 - val_accuracy: 0.6842\n",
      "Epoch 76/100\n",
      "6/6 [==============================] - 0s 19ms/step - loss: 4.4720 - accuracy: 0.6382 - val_loss: 2.3159 - val_accuracy: 0.6842\n",
      "Epoch 77/100\n",
      "6/6 [==============================] - 0s 18ms/step - loss: 4.6299 - accuracy: 0.6767 - val_loss: 2.4559 - val_accuracy: 0.7895\n",
      "Epoch 78/100\n",
      "6/6 [==============================] - 0s 19ms/step - loss: 4.9360 - accuracy: 0.6510 - val_loss: 2.9285 - val_accuracy: 0.7368\n",
      "Epoch 79/100\n",
      "6/6 [==============================] - 0s 19ms/step - loss: 3.3201 - accuracy: 0.7014 - val_loss: 3.3095 - val_accuracy: 0.7368\n",
      "Epoch 80/100\n",
      "6/6 [==============================] - 0s 19ms/step - loss: 4.3181 - accuracy: 0.6743 - val_loss: 3.3642 - val_accuracy: 0.7368\n",
      "Epoch 81/100\n",
      "6/6 [==============================] - 0s 18ms/step - loss: 3.9453 - accuracy: 0.7151 - val_loss: 3.3688 - val_accuracy: 0.7895\n",
      "Epoch 82/100\n",
      "6/6 [==============================] - 0s 19ms/step - loss: 3.0026 - accuracy: 0.7249 - val_loss: 3.0904 - val_accuracy: 0.7368\n",
      "Epoch 83/100\n",
      "6/6 [==============================] - 0s 19ms/step - loss: 3.3321 - accuracy: 0.7167 - val_loss: 2.8600 - val_accuracy: 0.7895\n",
      "Epoch 84/100\n",
      "6/6 [==============================] - 0s 19ms/step - loss: 3.1268 - accuracy: 0.7520 - val_loss: 2.8292 - val_accuracy: 0.7368\n",
      "Epoch 85/100\n",
      "6/6 [==============================] - 0s 20ms/step - loss: 3.1772 - accuracy: 0.7326 - val_loss: 2.4807 - val_accuracy: 0.7368\n",
      "Epoch 86/100\n",
      "6/6 [==============================] - 0s 17ms/step - loss: 2.8255 - accuracy: 0.7087 - val_loss: 2.3561 - val_accuracy: 0.6842\n",
      "Epoch 87/100\n",
      "6/6 [==============================] - 0s 14ms/step - loss: 3.7474 - accuracy: 0.6909 - val_loss: 2.1746 - val_accuracy: 0.7895\n",
      "Epoch 88/100\n",
      "6/6 [==============================] - 0s 15ms/step - loss: 3.3634 - accuracy: 0.6941 - val_loss: 2.0894 - val_accuracy: 0.7895\n",
      "Epoch 89/100\n",
      "6/6 [==============================] - 0s 16ms/step - loss: 2.8188 - accuracy: 0.7177 - val_loss: 2.4908 - val_accuracy: 0.5789\n",
      "Epoch 90/100\n",
      "6/6 [==============================] - 0s 17ms/step - loss: 3.4169 - accuracy: 0.6776 - val_loss: 2.6845 - val_accuracy: 0.6316\n",
      "Epoch 91/100\n",
      "6/6 [==============================] - 0s 18ms/step - loss: 3.2625 - accuracy: 0.6961 - val_loss: 2.4615 - val_accuracy: 0.6842\n",
      "Epoch 92/100\n",
      "6/6 [==============================] - 0s 17ms/step - loss: 2.8726 - accuracy: 0.7379 - val_loss: 2.0705 - val_accuracy: 0.7368\n",
      "Epoch 93/100\n",
      "6/6 [==============================] - 0s 17ms/step - loss: 2.6924 - accuracy: 0.7804 - val_loss: 1.9467 - val_accuracy: 0.7895\n",
      "Epoch 94/100\n",
      "6/6 [==============================] - 0s 19ms/step - loss: 2.7099 - accuracy: 0.7586 - val_loss: 1.9439 - val_accuracy: 0.7895\n",
      "Epoch 95/100\n",
      "6/6 [==============================] - 0s 19ms/step - loss: 3.4220 - accuracy: 0.6919 - val_loss: 2.0172 - val_accuracy: 0.7895\n",
      "Epoch 96/100\n",
      "6/6 [==============================] - 0s 19ms/step - loss: 2.9815 - accuracy: 0.7009 - val_loss: 2.1593 - val_accuracy: 0.6842\n",
      "Epoch 97/100\n",
      "6/6 [==============================] - 0s 18ms/step - loss: 3.1022 - accuracy: 0.7177 - val_loss: 2.3550 - val_accuracy: 0.6842\n",
      "Epoch 98/100\n",
      "6/6 [==============================] - 0s 20ms/step - loss: 2.3815 - accuracy: 0.8119 - val_loss: 2.5192 - val_accuracy: 0.7368\n",
      "Epoch 99/100\n",
      "6/6 [==============================] - 0s 17ms/step - loss: 2.0684 - accuracy: 0.7738 - val_loss: 2.5227 - val_accuracy: 0.6842\n",
      "Epoch 100/100\n",
      "6/6 [==============================] - 0s 18ms/step - loss: 2.9281 - accuracy: 0.7336 - val_loss: 2.5702 - val_accuracy: 0.6842\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "import json\n",
    "import numpy as np\n",
    "from sklearn.model_selection import train_test_split\n",
    "import tensorflow.keras as keras\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "# path to json file that stores MFCCs and genre labels for each processed segment\n",
    "DATA_PATH = \"data.json\"\n",
    "\n",
    "def load_data(data_path):\n",
    "    with open(data_path, \"r\") as fp:\n",
    "        data = json.load(fp)\n",
    "\n",
    "    # convert lists to numpy arrays\n",
    "    X = np.array(data[\"mfcc\"])\n",
    "    y = np.array(data[\"labels\"])\n",
    "\n",
    "    print(\"Data succesfully loaded!\")\n",
    "\n",
    "    return X, y\n",
    "\n",
    "\n",
    "def plot_history(history):\n",
    "    \"\"\"Plots accuracy/loss for training/validation set as a function of the epochs\n",
    "        :param history: Training history of model\n",
    "        :return:\n",
    "    \"\"\"\n",
    "\n",
    "    fig, axs = plt.subplots(2)\n",
    "\n",
    "    # create accuracy sublpot\n",
    "    axs[0].plot(history.history[\"accuracy\"], label=\"train accuracy\")\n",
    "    axs[0].plot(history.history[\"val_accuracy\"], label=\"test accuracy\")\n",
    "    axs[0].set_ylabel(\"Accuracy\")\n",
    "    axs[0].legend(loc=\"lower right\")\n",
    "    axs[0].set_title(\"Accuracy eval\")\n",
    "\n",
    "    # create error sublpot\n",
    "    axs[1].plot(history.history[\"loss\"], label=\"train error\")\n",
    "    axs[1].plot(history.history[\"val_loss\"], label=\"test error\")\n",
    "    axs[1].set_ylabel(\"Error\")\n",
    "    axs[1].set_xlabel(\"Epoch\")\n",
    "    axs[1].legend(loc=\"upper right\")\n",
    "    axs[1].set_title(\"Error eval\")\n",
    "\n",
    "    plt.show()\n",
    "\n",
    "\n",
    "if __name__ == \"__main__\":\n",
    "\n",
    "    # load data\n",
    "    X, y = load_data(DATA_PATH)\n",
    "\n",
    "    # create train/test split\n",
    "    X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.1)\n",
    "\n",
    "    # build network topology\n",
    "    model = keras.Sequential([\n",
    "\n",
    "        # input layer\n",
    "        keras.layers.Flatten(input_shape=(X.shape[1], X.shape[2])),\n",
    "\n",
    "        # 1st dense layer\n",
    "        keras.layers.Dense(512, activation='relu', kernel_regularizer=keras.regularizers.l2(0.001)),\n",
    "        keras.layers.Dropout(0.3),\n",
    "\n",
    "        # 2nd dense layer\n",
    "        keras.layers.Dense(256, activation='relu', kernel_regularizer=keras.regularizers.l2(0.001)),\n",
    "        keras.layers.Dropout(0.3),\n",
    "\n",
    "        # 3rd dense layer\n",
    "        keras.layers.Dense(64, activation='relu', kernel_regularizer=keras.regularizers.l2(0.001)),\n",
    "        keras.layers.Dropout(0.3),\n",
    "\n",
    "        # output layer\n",
    "        keras.layers.Dense(3, activation='softmax')\n",
    "    ])\n",
    "\n",
    "    # compile model\n",
    "    optimiser = keras.optimizers.Adam(learning_rate=0.0001)\n",
    "    model.compile(optimizer=optimiser,\n",
    "                  loss='sparse_categorical_crossentropy',\n",
    "                  metrics=['accuracy'])\n",
    "\n",
    "    model.summary()\n",
    "\n",
    "    # train model\n",
    "    history = model.fit(X_train, y_train, validation_data=(X_test, y_test), batch_size=32, epochs=100)\n",
    "\n",
    "    # plot accuracy and error as a function of the epochs\n",
    "    plot_history(history)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "b9349166",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential_5\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "conv2d_15 (Conv2D)           (None, 214, 11, 32)       320       \n",
      "_________________________________________________________________\n",
      "max_pooling2d_15 (MaxPooling (None, 107, 6, 32)        0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_15 (Batc (None, 107, 6, 32)        128       \n",
      "_________________________________________________________________\n",
      "conv2d_16 (Conv2D)           (None, 105, 4, 32)        9248      \n",
      "_________________________________________________________________\n",
      "max_pooling2d_16 (MaxPooling (None, 53, 2, 32)         0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_16 (Batc (None, 53, 2, 32)         128       \n",
      "_________________________________________________________________\n",
      "conv2d_17 (Conv2D)           (None, 52, 1, 32)         4128      \n",
      "_________________________________________________________________\n",
      "max_pooling2d_17 (MaxPooling (None, 26, 1, 32)         0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_17 (Batc (None, 26, 1, 32)         128       \n",
      "_________________________________________________________________\n",
      "flatten_5 (Flatten)          (None, 832)               0         \n",
      "_________________________________________________________________\n",
      "dense_10 (Dense)             (None, 64)                53312     \n",
      "_________________________________________________________________\n",
      "dropout_5 (Dropout)          (None, 64)                0         \n",
      "_________________________________________________________________\n",
      "dense_11 (Dense)             (None, 4)                 260       \n",
      "=================================================================\n",
      "Total params: 67,652\n",
      "Trainable params: 67,460\n",
      "Non-trainable params: 192\n",
      "_________________________________________________________________\n",
      "Epoch 1/110\n",
      "6/6 [==============================] - 1s 97ms/step - loss: 2.2589 - accuracy: 0.1877 - val_loss: 2.2440 - val_accuracy: 0.5556\n",
      "Epoch 2/110\n",
      "6/6 [==============================] - 0s 66ms/step - loss: 1.9448 - accuracy: 0.2976 - val_loss: 1.5459 - val_accuracy: 0.5556\n",
      "Epoch 3/110\n",
      "6/6 [==============================] - 0s 69ms/step - loss: 1.6825 - accuracy: 0.3313 - val_loss: 1.2102 - val_accuracy: 0.5556\n",
      "Epoch 4/110\n",
      "6/6 [==============================] - 0s 66ms/step - loss: 1.3167 - accuracy: 0.3911 - val_loss: 1.0097 - val_accuracy: 0.5556\n",
      "Epoch 5/110\n",
      "6/6 [==============================] - 0s 63ms/step - loss: 1.3972 - accuracy: 0.3953 - val_loss: 0.8703 - val_accuracy: 0.5556\n",
      "Epoch 6/110\n",
      "6/6 [==============================] - 0s 68ms/step - loss: 1.1835 - accuracy: 0.4980 - val_loss: 0.8303 - val_accuracy: 0.6667\n",
      "Epoch 7/110\n",
      "6/6 [==============================] - 0s 77ms/step - loss: 1.1166 - accuracy: 0.5534 - val_loss: 0.8261 - val_accuracy: 0.7778\n",
      "Epoch 8/110\n",
      "6/6 [==============================] - 0s 68ms/step - loss: 1.0199 - accuracy: 0.6081 - val_loss: 0.8365 - val_accuracy: 0.7778\n",
      "Epoch 9/110\n",
      "6/6 [==============================] - 0s 67ms/step - loss: 1.0437 - accuracy: 0.5901 - val_loss: 0.8308 - val_accuracy: 0.6667\n",
      "Epoch 10/110\n",
      "6/6 [==============================] - 0s 66ms/step - loss: 0.9639 - accuracy: 0.6369 - val_loss: 0.7964 - val_accuracy: 0.6667\n",
      "Epoch 11/110\n",
      "6/6 [==============================] - 0s 65ms/step - loss: 0.8372 - accuracy: 0.6593 - val_loss: 0.7491 - val_accuracy: 0.6667\n",
      "Epoch 12/110\n",
      "6/6 [==============================] - 0s 65ms/step - loss: 0.9351 - accuracy: 0.6122 - val_loss: 0.7149 - val_accuracy: 0.7778\n",
      "Epoch 13/110\n",
      "6/6 [==============================] - 0s 64ms/step - loss: 0.6893 - accuracy: 0.7490 - val_loss: 0.7047 - val_accuracy: 0.7778\n",
      "Epoch 14/110\n",
      "6/6 [==============================] - 0s 64ms/step - loss: 0.7313 - accuracy: 0.7136 - val_loss: 0.7088 - val_accuracy: 0.8889\n",
      "Epoch 15/110\n",
      "6/6 [==============================] - 0s 65ms/step - loss: 0.5743 - accuracy: 0.8000 - val_loss: 0.7086 - val_accuracy: 0.8889\n",
      "Epoch 16/110\n",
      "6/6 [==============================] - 0s 73ms/step - loss: 0.6156 - accuracy: 0.7749 - val_loss: 0.6980 - val_accuracy: 0.8889\n",
      "Epoch 17/110\n",
      "6/6 [==============================] - 0s 65ms/step - loss: 0.5923 - accuracy: 0.7967 - val_loss: 0.6826 - val_accuracy: 0.8889\n",
      "Epoch 18/110\n",
      "6/6 [==============================] - 0s 72ms/step - loss: 0.5284 - accuracy: 0.8077 - val_loss: 0.6694 - val_accuracy: 0.8889\n",
      "Epoch 19/110\n",
      "6/6 [==============================] - 0s 69ms/step - loss: 0.4532 - accuracy: 0.8655 - val_loss: 0.6607 - val_accuracy: 0.8889\n",
      "Epoch 20/110\n",
      "6/6 [==============================] - 0s 68ms/step - loss: 0.4367 - accuracy: 0.8599 - val_loss: 0.6562 - val_accuracy: 0.8889\n",
      "Epoch 21/110\n",
      "6/6 [==============================] - 0s 62ms/step - loss: 0.4526 - accuracy: 0.8703 - val_loss: 0.6599 - val_accuracy: 0.8889\n",
      "Epoch 22/110\n",
      "6/6 [==============================] - 0s 64ms/step - loss: 0.4931 - accuracy: 0.8168 - val_loss: 0.6633 - val_accuracy: 0.8889\n",
      "Epoch 23/110\n",
      "6/6 [==============================] - 0s 67ms/step - loss: 0.4699 - accuracy: 0.8502 - val_loss: 0.6582 - val_accuracy: 0.8889\n",
      "Epoch 24/110\n",
      "6/6 [==============================] - 1s 89ms/step - loss: 0.4097 - accuracy: 0.8424 - val_loss: 0.6406 - val_accuracy: 0.8889\n",
      "Epoch 25/110\n",
      "6/6 [==============================] - 0s 74ms/step - loss: 0.4193 - accuracy: 0.8807 - val_loss: 0.6180 - val_accuracy: 0.8889\n",
      "Epoch 26/110\n",
      "6/6 [==============================] - 0s 77ms/step - loss: 0.4097 - accuracy: 0.8467 - val_loss: 0.6088 - val_accuracy: 0.7778\n",
      "Epoch 27/110\n",
      "6/6 [==============================] - 0s 79ms/step - loss: 0.4020 - accuracy: 0.8789 - val_loss: 0.5932 - val_accuracy: 0.7778\n",
      "Epoch 28/110\n",
      "6/6 [==============================] - 1s 96ms/step - loss: 0.3217 - accuracy: 0.9068 - val_loss: 0.6099 - val_accuracy: 0.7778\n",
      "Epoch 29/110\n",
      "6/6 [==============================] - 1s 102ms/step - loss: 0.3583 - accuracy: 0.8841 - val_loss: 0.6402 - val_accuracy: 0.7778\n",
      "Epoch 30/110\n",
      "6/6 [==============================] - 0s 77ms/step - loss: 0.3537 - accuracy: 0.8786 - val_loss: 0.6449 - val_accuracy: 0.7778\n",
      "Epoch 31/110\n",
      "6/6 [==============================] - 1s 97ms/step - loss: 0.3812 - accuracy: 0.8885 - val_loss: 0.6281 - val_accuracy: 0.7778\n",
      "Epoch 32/110\n",
      "6/6 [==============================] - 1s 120ms/step - loss: 0.3378 - accuracy: 0.8755 - val_loss: 0.6505 - val_accuracy: 0.7778\n",
      "Epoch 33/110\n",
      "6/6 [==============================] - 1s 89ms/step - loss: 0.3039 - accuracy: 0.8897 - val_loss: 0.6623 - val_accuracy: 0.7778\n",
      "Epoch 34/110\n",
      "6/6 [==============================] - 0s 63ms/step - loss: 0.3037 - accuracy: 0.9086 - val_loss: 0.6800 - val_accuracy: 0.5556\n",
      "Epoch 35/110\n",
      "6/6 [==============================] - 0s 63ms/step - loss: 0.2840 - accuracy: 0.9319 - val_loss: 0.6535 - val_accuracy: 0.5556\n",
      "Epoch 36/110\n",
      "6/6 [==============================] - 0s 63ms/step - loss: 0.3190 - accuracy: 0.9172 - val_loss: 0.6153 - val_accuracy: 0.7778\n",
      "Epoch 37/110\n",
      "6/6 [==============================] - 0s 67ms/step - loss: 0.2626 - accuracy: 0.9382 - val_loss: 0.6006 - val_accuracy: 0.7778\n",
      "Epoch 38/110\n",
      "6/6 [==============================] - 0s 64ms/step - loss: 0.2954 - accuracy: 0.9312 - val_loss: 0.5684 - val_accuracy: 0.7778\n",
      "Epoch 39/110\n",
      "6/6 [==============================] - 0s 64ms/step - loss: 0.2786 - accuracy: 0.9002 - val_loss: 0.5418 - val_accuracy: 0.7778\n",
      "Epoch 40/110\n",
      "6/6 [==============================] - 0s 64ms/step - loss: 0.2564 - accuracy: 0.9483 - val_loss: 0.5036 - val_accuracy: 0.8889\n",
      "Epoch 41/110\n",
      "6/6 [==============================] - 0s 61ms/step - loss: 0.2078 - accuracy: 0.9672 - val_loss: 0.4609 - val_accuracy: 0.8889\n",
      "Epoch 42/110\n",
      "6/6 [==============================] - 0s 62ms/step - loss: 0.4230 - accuracy: 0.8303 - val_loss: 0.4324 - val_accuracy: 0.8889\n",
      "Epoch 43/110\n",
      "6/6 [==============================] - 0s 60ms/step - loss: 0.1940 - accuracy: 0.9750 - val_loss: 0.4214 - val_accuracy: 0.8889\n",
      "Epoch 44/110\n",
      "6/6 [==============================] - 0s 63ms/step - loss: 0.2287 - accuracy: 0.9539 - val_loss: 0.4213 - val_accuracy: 0.8889\n",
      "Epoch 45/110\n",
      "6/6 [==============================] - 0s 68ms/step - loss: 0.1934 - accuracy: 0.9417 - val_loss: 0.4359 - val_accuracy: 0.8889\n",
      "Epoch 46/110\n",
      "6/6 [==============================] - 0s 64ms/step - loss: 0.1928 - accuracy: 0.9689 - val_loss: 0.4526 - val_accuracy: 0.8889\n",
      "Epoch 47/110\n",
      "6/6 [==============================] - 0s 64ms/step - loss: 0.2029 - accuracy: 0.9487 - val_loss: 0.4936 - val_accuracy: 0.8889\n",
      "Epoch 48/110\n",
      "6/6 [==============================] - 0s 66ms/step - loss: 0.1970 - accuracy: 0.9276 - val_loss: 0.5225 - val_accuracy: 0.8889\n",
      "Epoch 49/110\n",
      "6/6 [==============================] - 0s 66ms/step - loss: 0.2018 - accuracy: 0.9579 - val_loss: 0.5455 - val_accuracy: 0.7778\n",
      "Epoch 50/110\n",
      "6/6 [==============================] - 0s 71ms/step - loss: 0.1983 - accuracy: 0.9502 - val_loss: 0.5368 - val_accuracy: 0.7778\n",
      "Epoch 51/110\n",
      "6/6 [==============================] - 0s 64ms/step - loss: 0.2122 - accuracy: 0.9728 - val_loss: 0.5123 - val_accuracy: 0.7778\n",
      "Epoch 52/110\n",
      "6/6 [==============================] - 0s 69ms/step - loss: 0.1849 - accuracy: 0.9834 - val_loss: 0.4865 - val_accuracy: 0.7778\n",
      "Epoch 53/110\n",
      "6/6 [==============================] - 0s 65ms/step - loss: 0.2007 - accuracy: 0.9631 - val_loss: 0.4711 - val_accuracy: 0.7778\n",
      "Epoch 54/110\n",
      "6/6 [==============================] - 0s 62ms/step - loss: 0.1729 - accuracy: 0.9595 - val_loss: 0.4617 - val_accuracy: 0.8889\n",
      "Epoch 55/110\n",
      "6/6 [==============================] - 0s 69ms/step - loss: 0.1669 - accuracy: 0.9728 - val_loss: 0.4759 - val_accuracy: 0.8889\n",
      "Epoch 56/110\n",
      "6/6 [==============================] - 0s 65ms/step - loss: 0.1885 - accuracy: 0.9663 - val_loss: 0.4711 - val_accuracy: 0.8889\n",
      "Epoch 57/110\n",
      "6/6 [==============================] - 0s 64ms/step - loss: 0.1881 - accuracy: 0.9594 - val_loss: 0.4598 - val_accuracy: 0.8889\n",
      "Epoch 58/110\n",
      "6/6 [==============================] - 0s 65ms/step - loss: 0.2166 - accuracy: 0.9531 - val_loss: 0.4997 - val_accuracy: 0.6667\n",
      "Epoch 59/110\n",
      "6/6 [==============================] - 0s 63ms/step - loss: 0.1454 - accuracy: 0.9815 - val_loss: 0.5247 - val_accuracy: 0.6667\n",
      "Epoch 60/110\n",
      "6/6 [==============================] - 0s 57ms/step - loss: 0.1830 - accuracy: 0.9343 - val_loss: 0.6054 - val_accuracy: 0.5556\n",
      "Epoch 61/110\n",
      "6/6 [==============================] - 0s 60ms/step - loss: 0.1513 - accuracy: 0.9598 - val_loss: 0.5925 - val_accuracy: 0.5556\n",
      "Epoch 62/110\n",
      "6/6 [==============================] - 0s 62ms/step - loss: 0.1506 - accuracy: 0.9628 - val_loss: 0.5626 - val_accuracy: 0.5556\n",
      "Epoch 63/110\n",
      "6/6 [==============================] - 0s 62ms/step - loss: 0.1477 - accuracy: 0.9665 - val_loss: 0.5333 - val_accuracy: 0.5556\n",
      "Epoch 64/110\n",
      "6/6 [==============================] - 0s 64ms/step - loss: 0.1423 - accuracy: 0.9807 - val_loss: 0.5260 - val_accuracy: 0.5556\n",
      "Epoch 65/110\n",
      "6/6 [==============================] - 0s 70ms/step - loss: 0.1417 - accuracy: 0.9802 - val_loss: 0.4940 - val_accuracy: 0.6667\n",
      "Epoch 66/110\n",
      "6/6 [==============================] - 0s 59ms/step - loss: 0.1529 - accuracy: 0.9624 - val_loss: 0.4071 - val_accuracy: 0.6667\n",
      "Epoch 67/110\n",
      "6/6 [==============================] - 0s 81ms/step - loss: 0.1145 - accuracy: 0.9765 - val_loss: 0.3843 - val_accuracy: 0.6667\n",
      "Epoch 68/110\n",
      "6/6 [==============================] - 0s 63ms/step - loss: 0.1684 - accuracy: 0.9230 - val_loss: 0.3921 - val_accuracy: 0.6667\n",
      "Epoch 69/110\n",
      "6/6 [==============================] - 0s 61ms/step - loss: 0.1492 - accuracy: 0.9563 - val_loss: 0.4014 - val_accuracy: 0.6667\n",
      "Epoch 70/110\n",
      "6/6 [==============================] - 0s 64ms/step - loss: 0.1478 - accuracy: 0.9555 - val_loss: 0.3984 - val_accuracy: 0.6667\n",
      "Epoch 71/110\n",
      "6/6 [==============================] - 0s 70ms/step - loss: 0.1190 - accuracy: 0.9894 - val_loss: 0.3995 - val_accuracy: 0.7778\n",
      "Epoch 72/110\n",
      "6/6 [==============================] - 0s 68ms/step - loss: 0.1100 - accuracy: 0.9874 - val_loss: 0.3935 - val_accuracy: 0.7778\n",
      "Epoch 73/110\n",
      "6/6 [==============================] - 0s 60ms/step - loss: 0.1216 - accuracy: 0.9874 - val_loss: 0.4047 - val_accuracy: 0.7778\n",
      "Epoch 74/110\n",
      "6/6 [==============================] - 0s 59ms/step - loss: 0.1157 - accuracy: 0.9847 - val_loss: 0.4086 - val_accuracy: 0.7778\n",
      "Epoch 75/110\n",
      "6/6 [==============================] - 0s 59ms/step - loss: 0.1415 - accuracy: 0.9551 - val_loss: 0.4007 - val_accuracy: 0.7778\n",
      "Epoch 76/110\n",
      "6/6 [==============================] - 0s 59ms/step - loss: 0.0993 - accuracy: 0.9948 - val_loss: 0.4028 - val_accuracy: 0.6667\n",
      "Epoch 77/110\n",
      "6/6 [==============================] - 0s 57ms/step - loss: 0.1234 - accuracy: 0.9785 - val_loss: 0.3641 - val_accuracy: 0.6667\n",
      "Epoch 78/110\n",
      "6/6 [==============================] - 0s 63ms/step - loss: 0.0991 - accuracy: 0.9754 - val_loss: 0.2928 - val_accuracy: 0.7778\n",
      "Epoch 79/110\n",
      "6/6 [==============================] - 0s 58ms/step - loss: 0.1090 - accuracy: 0.9900 - val_loss: 0.2861 - val_accuracy: 0.8889\n",
      "Epoch 80/110\n",
      "6/6 [==============================] - 0s 63ms/step - loss: 0.0903 - accuracy: 0.9948 - val_loss: 0.2904 - val_accuracy: 0.8889\n",
      "Epoch 81/110\n",
      "6/6 [==============================] - 0s 62ms/step - loss: 0.1230 - accuracy: 0.9693 - val_loss: 0.2899 - val_accuracy: 0.8889\n",
      "Epoch 82/110\n",
      "6/6 [==============================] - 0s 70ms/step - loss: 0.1113 - accuracy: 0.9931 - val_loss: 0.3099 - val_accuracy: 0.7778\n",
      "Epoch 83/110\n",
      "6/6 [==============================] - 0s 63ms/step - loss: 0.0811 - accuracy: 0.9948 - val_loss: 0.3167 - val_accuracy: 0.8889\n",
      "Epoch 84/110\n",
      "6/6 [==============================] - 0s 64ms/step - loss: 0.0865 - accuracy: 0.9855 - val_loss: 0.3700 - val_accuracy: 0.8889\n",
      "Epoch 85/110\n",
      "6/6 [==============================] - 0s 60ms/step - loss: 0.0780 - accuracy: 0.9911 - val_loss: 0.4078 - val_accuracy: 0.8889\n",
      "Epoch 86/110\n",
      "6/6 [==============================] - 0s 61ms/step - loss: 0.0910 - accuracy: 0.9926 - val_loss: 0.4101 - val_accuracy: 0.8889\n",
      "Epoch 87/110\n",
      "6/6 [==============================] - 0s 62ms/step - loss: 0.0787 - accuracy: 0.9974 - val_loss: 0.3957 - val_accuracy: 0.7778\n",
      "Epoch 88/110\n",
      "6/6 [==============================] - 0s 61ms/step - loss: 0.1497 - accuracy: 0.9542 - val_loss: 0.3535 - val_accuracy: 0.7778\n",
      "Epoch 89/110\n",
      "6/6 [==============================] - 0s 62ms/step - loss: 0.1157 - accuracy: 0.9851 - val_loss: 0.4111 - val_accuracy: 0.7778\n",
      "Epoch 90/110\n",
      "6/6 [==============================] - 0s 59ms/step - loss: 0.0805 - accuracy: 0.9937 - val_loss: 0.4398 - val_accuracy: 0.7778\n",
      "Epoch 91/110\n",
      "6/6 [==============================] - 0s 63ms/step - loss: 0.0719 - accuracy: 1.0000 - val_loss: 0.4112 - val_accuracy: 0.7778\n",
      "Epoch 92/110\n",
      "6/6 [==============================] - 0s 59ms/step - loss: 0.0734 - accuracy: 0.9896 - val_loss: 0.3801 - val_accuracy: 0.8889\n",
      "Epoch 93/110\n",
      "6/6 [==============================] - 0s 61ms/step - loss: 0.0879 - accuracy: 0.9818 - val_loss: 0.3218 - val_accuracy: 0.8889\n",
      "Epoch 94/110\n",
      "6/6 [==============================] - 0s 62ms/step - loss: 0.1055 - accuracy: 0.9851 - val_loss: 0.3265 - val_accuracy: 0.8889\n",
      "Epoch 95/110\n",
      "6/6 [==============================] - 0s 61ms/step - loss: 0.0828 - accuracy: 0.9900 - val_loss: 0.3339 - val_accuracy: 0.8889\n",
      "Epoch 96/110\n",
      "6/6 [==============================] - 0s 58ms/step - loss: 0.0998 - accuracy: 0.9885 - val_loss: 0.3446 - val_accuracy: 0.8889\n",
      "Epoch 97/110\n",
      "6/6 [==============================] - 0s 60ms/step - loss: 0.0971 - accuracy: 0.9894 - val_loss: 0.3618 - val_accuracy: 0.7778\n",
      "Epoch 98/110\n",
      "6/6 [==============================] - 0s 59ms/step - loss: 0.0853 - accuracy: 0.9834 - val_loss: 0.3958 - val_accuracy: 0.7778\n",
      "Epoch 99/110\n",
      "6/6 [==============================] - 0s 62ms/step - loss: 0.0939 - accuracy: 0.9926 - val_loss: 0.4113 - val_accuracy: 0.8889\n",
      "Epoch 100/110\n",
      "6/6 [==============================] - 0s 59ms/step - loss: 0.0675 - accuracy: 0.9948 - val_loss: 0.4140 - val_accuracy: 0.8889\n",
      "Epoch 101/110\n",
      "6/6 [==============================] - 0s 60ms/step - loss: 0.0635 - accuracy: 0.9948 - val_loss: 0.4539 - val_accuracy: 0.8889\n",
      "Epoch 102/110\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "6/6 [==============================] - 0s 61ms/step - loss: 0.0865 - accuracy: 0.9881 - val_loss: 0.4019 - val_accuracy: 0.8889\n",
      "Epoch 103/110\n",
      "6/6 [==============================] - 0s 67ms/step - loss: 0.0691 - accuracy: 0.9948 - val_loss: 0.3727 - val_accuracy: 0.8889\n",
      "Epoch 104/110\n",
      "6/6 [==============================] - 1s 75ms/step - loss: 0.1190 - accuracy: 0.9746 - val_loss: 0.4016 - val_accuracy: 0.8889\n",
      "Epoch 105/110\n",
      "6/6 [==============================] - 1s 108ms/step - loss: 0.0583 - accuracy: 1.0000 - val_loss: 0.5024 - val_accuracy: 0.8889\n",
      "Epoch 106/110\n",
      "6/6 [==============================] - 0s 79ms/step - loss: 0.0597 - accuracy: 0.9966 - val_loss: 0.5039 - val_accuracy: 0.7778\n",
      "Epoch 107/110\n",
      "6/6 [==============================] - 0s 62ms/step - loss: 0.0577 - accuracy: 0.9926 - val_loss: 0.4179 - val_accuracy: 0.7778\n",
      "Epoch 108/110\n",
      "6/6 [==============================] - 0s 59ms/step - loss: 0.0734 - accuracy: 0.9974 - val_loss: 0.3536 - val_accuracy: 0.7778\n",
      "Epoch 109/110\n",
      "6/6 [==============================] - 0s 64ms/step - loss: 0.0615 - accuracy: 0.9946 - val_loss: 0.3247 - val_accuracy: 0.6667\n",
      "Epoch 110/110\n",
      "6/6 [==============================] - 0s 61ms/step - loss: 0.0720 - accuracy: 0.9881 - val_loss: 0.3102 - val_accuracy: 0.7778\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1/1 - 0s - loss: 0.8350 - accuracy: 0.8000\n",
      "\n",
      "Test accuracy: 0.800000011920929\n",
      "WARNING:tensorflow:6 out of the last 6 calls to <function Model.make_predict_function.<locals>.predict_function at 0x155db8670> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings could be due to (1) creating @tf.function repeatedly in a loop, (2) passing tensors with different shapes, (3) passing Python objects instead of tensors. For (1), please define your @tf.function outside of the loop. For (2), @tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. For (3), please refer to https://www.tensorflow.org/guide/function#controlling_retracing and https://www.tensorflow.org/api_docs/python/tf/function for  more details.\n",
      "Target: 0, Predicted label: [0]\n"
     ]
    }
   ],
   "source": [
    "import json\n",
    "import numpy as np\n",
    "from sklearn.model_selection import train_test_split\n",
    "import tensorflow.keras as keras\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "DATA_PATH = \"data.json\"\n",
    "\n",
    "\n",
    "def load_data(data_path):\n",
    "    \"\"\"Loads training dataset from json file.\n",
    "        :param data_path (str): Path to json file containing data\n",
    "        :return X (ndarray): Inputs\n",
    "        :return y (ndarray): Targets\n",
    "    \"\"\"\n",
    "\n",
    "    with open(data_path, \"r\") as fp:\n",
    "        data = json.load(fp)\n",
    "\n",
    "    X = np.array(data[\"mfcc\"])\n",
    "    y = np.array(data[\"labels\"])\n",
    "    return X, y\n",
    "\n",
    "\n",
    "def plot_history(history):\n",
    "    \"\"\"Plots accuracy/loss for training/validation set as a function of the epochs\n",
    "        :param history: Training history of model\n",
    "        :return:\n",
    "    \"\"\"\n",
    "\n",
    "    fig, axs = plt.subplots(2)\n",
    "\n",
    "    # create accuracy sublpot\n",
    "    axs[0].plot(history.history[\"accuracy\"], label=\"train accuracy\")\n",
    "    axs[0].plot(history.history[\"val_accuracy\"], label=\"test accuracy\")\n",
    "    axs[0].set_ylabel(\"Accuracy\")\n",
    "    axs[0].legend(loc=\"lower right\")\n",
    "    axs[0].set_title(\"Accuracy eval\")\n",
    "\n",
    "    # create error sublpot\n",
    "    axs[1].plot(history.history[\"loss\"], label=\"train error\")\n",
    "    axs[1].plot(history.history[\"val_loss\"], label=\"test error\")\n",
    "    axs[1].set_ylabel(\"Error\")\n",
    "    axs[1].set_xlabel(\"Epoch\")\n",
    "    axs[1].legend(loc=\"upper right\")\n",
    "    axs[1].set_title(\"Error eval\")\n",
    "\n",
    "    plt.show()\n",
    "\n",
    "def prepare_datasets(test_size, validation_size):\n",
    "    # load data\n",
    "    X, y = load_data(DATA_PATH)\n",
    "\n",
    "    # create train, validation and test split\n",
    "    X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=test_size)\n",
    "    X_train, X_validation, y_train, y_validation = train_test_split(X_train, y_train, test_size=validation_size)\n",
    "\n",
    "    # add an axis to input sets\n",
    "    X_train = X_train[..., np.newaxis]\n",
    "    X_validation = X_validation[..., np.newaxis]\n",
    "    X_test = X_test[..., np.newaxis]\n",
    "\n",
    "    return X_train, X_validation, X_test, y_train, y_validation, y_test\n",
    "\n",
    "def build_model(input_shape):\n",
    "    \n",
    "    # build network topology\n",
    "    model = keras.Sequential()\n",
    "\n",
    "    # 1st conv layer\n",
    "    model.add(keras.layers.Conv2D(32, (3, 3), activation='relu', input_shape=input_shape))\n",
    "    model.add(keras.layers.MaxPooling2D((3, 3), strides=(2, 2), padding='same'))\n",
    "    model.add(keras.layers.BatchNormalization())\n",
    "\n",
    "    # 2nd conv layer\n",
    "    model.add(keras.layers.Conv2D(32, (3, 3), activation='relu'))\n",
    "    model.add(keras.layers.MaxPooling2D((3, 3), strides=(2, 2), padding='same'))\n",
    "    model.add(keras.layers.BatchNormalization())\n",
    "\n",
    "    # 3rd conv layer\n",
    "    model.add(keras.layers.Conv2D(32, (2, 2), activation='relu'))\n",
    "    model.add(keras.layers.MaxPooling2D((2, 2), strides=(2, 2), padding='same'))\n",
    "    model.add(keras.layers.BatchNormalization())\n",
    "\n",
    "    # flatten output and feed it into dense layer\n",
    "    model.add(keras.layers.Flatten())\n",
    "    model.add(keras.layers.Dense(64, activation='relu'))\n",
    "    model.add(keras.layers.Dropout(0.2))\n",
    "\n",
    "    # output layer\n",
    "    model.add(keras.layers.Dense(4, activation='softmax'))\n",
    "\n",
    "    return model\n",
    "\n",
    "\n",
    "def predict(model, X, y):\n",
    "    \"\"\"Predict a single sample using the trained model\n",
    "    :param model: Trained classifier\n",
    "    :param X: Input data\n",
    "    :param y (int): Target\n",
    "    \"\"\"\n",
    "\n",
    "    # add a dimension to input data for sample - model.predict() expects a 4d array in this case\n",
    "    X = X[np.newaxis, ...] # array shape (1, 130, 13, 1)\n",
    "\n",
    "    # perform prediction\n",
    "    prediction = model.predict(X)\n",
    "\n",
    "    # get index with max value\n",
    "    predicted_index = np.argmax(prediction, axis=1)\n",
    "\n",
    "    print(\"Target: {}, Predicted label: {}\".format(y, predicted_index))\n",
    "\n",
    "\n",
    "if __name__ == \"__main__\":\n",
    "\n",
    "    # get train, validation, test splits\n",
    "    X_train, X_validation, X_test, y_train, y_validation, y_test = prepare_datasets(0.05, 0.05)\n",
    "    # create network\n",
    "    input_shape = (X_train.shape[1], X_train.shape[2], 1)\n",
    "    model = build_model(input_shape)\n",
    "\n",
    "    # compile model\n",
    "    optimiser = keras.optimizers.Adam(learning_rate=0.0001)\n",
    "    model.compile(optimizer=optimiser,\n",
    "                  loss='sparse_categorical_crossentropy',\n",
    "                  metrics=['accuracy'])\n",
    "\n",
    "    model.summary()\n",
    "\n",
    "    # train model\n",
    "#     X_train = X_train / 255\n",
    "    history = model.fit(X_train, y_train, validation_data=(X_validation, y_validation), batch_size=32, epochs=110)\n",
    "\n",
    "    # plot accuracy/error for training and validation\n",
    "    plot_history(history)\n",
    "\n",
    "    # evaluate model on test set\n",
    "    test_loss, test_acc = model.evaluate(X_test, y_test, verbose=2)\n",
    "    print('\\nTest accuracy:', test_acc)\n",
    "\n",
    "    # pick a sample to predict from the test set\n",
    "    X_to_predict = X_test[1]\n",
    "    y_to_predict = y_test[1]\n",
    "\n",
    "    # predict sample\n",
    "    predict(model, X_to_predict, y_to_predict)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "04f6e22e",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Save the model as a file\n",
    "from tensorflow.keras import models\n",
    "models.save_model(model, 'final-model.h5')"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3.8 (tensorflow)",
   "language": "python",
   "name": "tensorflow"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.10"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}