example_multi_spectrogram_classificaiton_phoneme-sliding.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Example: phoneme classification with multiple samples spectrogram \n",
    "Experimental notebook by @ste\n",
    "\n",
    "TL;DR\n",
    "\n",
    "The current dataset has a lot of short words and some long words.\n",
    "If you look at the samples of \"single spectrogram\" notebook you'll see that for short sounds almost all the resulting spectrogram is \"silence\".\n",
    "\n",
    "So I've tried to create a sample notebook that, instead of a single channel, transform the \"augmented\" sound (volume, noise, ...) into a three channel image where:\n",
    " - channel 0: is the complete spectrogram (like single spectrogram notebook)\n",
    " - channel 1: contains a spectrogram created with the first 50% of the sound (for medium lenght sounds)\n",
    " - channel 2: is the spectrogram for short sound \n",
    " \n",
    "**With the single spectrogram approach (using 10% of train/valid) I've got 29% accouracy, with multi spectrogram approach I've got 37%.**"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The key point is understanding the \"chain of transformations\" that takes the original data (a \"sound\") and converts them it into an suitable input for the model.\n",
    "This chain comprises two main parts:\n",
    "* DATA AUGMENTATION: f(sound) -> sound \n",
    "* STRUCTURE TRANSFORM: f(sound) -> tensor"
   ]
  },
  {
   "attachments": {
    "image.png": {
     "image/png": "iVBORw0KGgoAAAANSUhEUgAAAlgAAABcCAYAAACyaMvjAAAgAElEQVR4Ae3dCZheRZU38NN7d9KdjRB2BFQ2wRFwxX0bR8cdB5lRBxdcxlFUxtHRwW1mFMUd3FFG1HFwwRV0FP0AURFUEBUEUbYAJiQhe3rv/p5fdSq5eek13Um6O1XPc99733tru/+qOvWvU6fq1g0ODg5GcQWBgkBBoCBQECgIFAQKAlOGQP2UxVQiKggUBAoCBYGCQEGgIFAQSAgUglUqQkGgIFAQKAgUBAoCBYEpRqAQrCkGtERXECgIFAQKAgWBgkBBoBCsUgcKAgWBgkBBoCBQECgITDEChWBNMaAluoJAQaAgUBAoCBQECgKFYJU6UBAoCBQECgIFgYJAQWCKESgEa4oBLdEVBAoCBYGCQEGgIFAQKASr1IGCQEGgIFAQKAgUBAoCU4xA4xTHV6IrCBQECgLjRsA+x466urp0jCdgf39/9Pf1RX1DQzQ2FhE2HsyKn4JAQWDnI1A0WDsf85JiQaAgsBmB3t7e+M53vhP33HNPIlrjAeZXv/pVPP3pT49LfvS9GBjoH0+Q4qcgUBAoCOx0BArB2umQlwQLAgWBjMDdd98d73rnO+K0094Q69evz7dHPA8MDMRPL/9J3HbDFfH2f315/N/3Loy+vr4R/ZcHBYGCQEFgVyFQCNauQr6kWxCYpQiY8vvZz34W55xzTiJNSNFIrq9rTRxzYF/cf96yeOlLTo5Vq1aNqsnq7u6Oq678ebzpxAPii2/YN847683x/jPfG52dnSMlsc19ZOz3v/993HjjjdvcL38KArsDAlljfO6554br4nYsAsWAYcfiW2IvCOx2CAwODMQHP/CBuO3GK+Lr558Xz3v+i+JF/3hyNDc3R339tmO6wb6uaKrrihc+vDPmXtsUbzzt1Dj97e+Kgw46OBoaGu6FHWK0+i83xjOeNz/mNPXHGS9cGP91wVfinM/MiVe+6p+jpaXlXmHyDbZbF114Ybzu9a+LJz/pSXHOOZ+NqKvLj8u5IDDrEdi0aVN88IMfjOuvvz7+/Oc/x+GHHx5LliyJtra2aGpqSjaN2h2bSG01t0GDJte5DbvO/t3L/oTjsk1lPud7I52zv2r4FFFEGnAZGMlPrb/sZ7qeC8GariVT8lUQmKEI1NXXx8EH7R9z1kW84YS+eOtn3x6/ufqX8W9vPT322We/aGxq2vJm9Q3N0dLUEO2NnXHyw/riy7/si396xUvjI2d/Mu5//0OTEM+eCdlPf+qT8cQHzo22poHEjfbq6Iv3vGBRvOJTn4/DDj8invzkp9ybxA0OBnL1k59cFqe99pQ4ZM+6mNc2mKMt54LAboPA3Llz47zzzks2jB/60IfSgOR+97tfdHR0bCFPFo7kgRDilElNBgnZyoSHP22LH/aQfb29iawhXXPb22NeR0c0NTdHc1PT5nNz1NXXpf8NDY3R2NQYc+e2R2trS8yZMyfmdcyL5paWlK+crni//vULYs8990zHIx7xiDjooIO25DHnazqe6wahVVxBoCBQEJhCBG696fdx0glPia+9cUlQWn3qohVx0XXt8cEPfSge9ZgnRGNTc9TV1cedd9weZ/zLM+Kdz2mMtpaGGKhvjh9f3xinn3d7fORjn47HPe7xW1YKLl26NE76u2fF51+3X+zbuqyS27q4/PrOeP+P5sU3vvW9mD9//pZnuTO47LJL402vf3mc+YK58akfrInXvfPcePTjnrLFX7koCOwuCGgTK1eujFe96lXx61//Oi666KJEWJCiTGqcs8v3hDPdXz27RrBMNy5fdmf89porgwY7ubq69Mzzvr7+FNb9IZOBuujr642B/v4YGIx03dnVFd3dPYm8dXV1R29vXwwMDkRPT0987/sXx7LlK1LYl7zkJfGJT3wiEcKcx+l6bnjnO9/5zumauZKvgkBBYGYi0DF/QfR0borPnf+jOOGRi+L4B7THvKYN8a4Pnx+r774tjjv6vtE8d4/o6u6J73/na/HEIweiubEu6gb7476Le+N+S3rjP8/+drR27BlHHXVUGq1+4QtfiPvN/XM8+pB10VC37bhwn4UNcdMtd8dvb+uNxzzmMVtAI9wvvPA7ceY7XhNnvrA97rOkOX618pB4yav+dUYI6C0vUi4KAlOEAMJEk3Xd738fN930x/jD9dfHscceF3vttVdqZ7VTcZlsOXvmoNmi6XIgZq2trdHR3hFrVq+I+vq6aG/viKbGhmhrbY05c9qivX1udHS0p2PevI6YN689FiyYHwsXLohFixbE4sV7xD577xX777dvHHjA/nHIwfeJ+9/v4Dj0fofE4YfeLx71yIcl/zfffGvK5xOe+ISUxhRBssOi2dYgYoclUyIuCBQEdicE6hua4sS/f1HcsHJOXPrbdUHQPPf4RfHpV+8dX/vKl+PZz3leXP/rH8fgQG90DrTEYP3WaUM4PeaoefG2E1rj3W8/Lb72ta/FmjVr4mNnfSSeeOze0Vi/Lbniv6mhLk56dEd87pMfjGuuuSaNdK1K/OQnzoqP/ser49OvXBDHHNIWV/y5MR71pBNSh7A7lUd514JAFQGap7ntcxOJecJjj4+Tnn9iXHrppUnjVPU3kWtTewfc575h6q+3t2ciQUf1i9gtXLAgnv2Mp8bHPvLeuPPO2+OEE06Im266adRw0+FhmSKcDqVQ8lAQmIUI9PX1xA2//mG84Z9fGue/6YBoaayP/oHBWL6mN9739WVx2Q0Rj3nUI+Oa3/w2LnjLgbHXnG23aejtG4zLr9sQp56zPBbve99o6FsT3/r3g6OjYS3T13sh1t07EO/92vK4qfe4eM8Z7413vePf48iOm+LkxzTFwvaG6OuPOOEDK+Ojn/1WHH30A9N0yL0iKTcKArsJAmef9ZH42eWXxd8//zmxctU9cfYnzo39DzgwDU7YZDksGmEbheTQUmVtFqP4PF3I2N1904TurV97T6xevWrITovWq2HITmtwYMhQnv/+gf6or6uPpqYhLVhDff2QKYANhyOit68vHvrgY5LGq1oc4t/U2RV/vvWu+O6F34/TTjst2ZPlPFT9TofrYuQ+HUqh5KEgMAsRaGxsjkMf9KR44CNPiE9/73vxT3+7Z5oG3GdRc5z5kv3j/MtWxWd/+MNYu2kg1ncdHHvNIVq3Eqemxrp49APa46yXD8SFVy2N5z710dHWcNc2fqqwNTfWx2nP2StOePelccKznhr/ctKh8eyj2qK1ie1IxFU3bYoF+xwRBx98yJaOohp+Nl7TVOROcTa+32TeCTZVl/87V48NGzbEtddem4jGgx/84C0r5qphZ+L1okWLoqm5Kb3P4j32SFqnq6++OhGlZz7zmbHPPvtEe3t7em9TgsiWM4fQmCqEk2lCdQz5ydfsqzZuWB/r161Jz0zVe85gne2lsP39fck2i9/+vv6obxgyrq+rb4i//GVZdMxfFB3tbaGYmpqGViAPDg6kLzg87olPi+Mf+Zg4++yz4+abbw52WQsXLkzxTqeyKBqs6VQaJS8FgVmGAKFqNPvMJz80vvjahbFkvlVJQy9JcG7oYuQ6ZBPSGN0RgyPvmRVpbCvsth1jFbKBgcFY3zUQfc37xvymDdE4uCn57+0fjLd8fln841vPj4c+7BFbOopq2Nlw3dXVtWWaR+e3bNmypImghcj2M1XCVb2ufX/hueq5el3rv+o3X+f4c7hqmOHuVZ+P91o8OZ2cbm1YHTw/MLCXWk7bajh11Bl2yJQ91Zx9XWD16tXxgx/8IK644oqE6+WXXx5HH330NunVpjVT/n/jgq/HRd/9ZjznWX+bMPjAhz8erz/tjfGZz5yT3vFtb3tbWhEItyq+tRjDMuOZ3z3/R4i0Sa4JEavX+MW3tRkPbm7PQ+kMbcUwFKewjOoHor5+aDVj8loX6T8/yuw///M/45ZbbomPfOQjsXjx4iFNWM7ILj4XDdYuLoCSfEFgNiOgQ1u4cI94+WtPjy9e+p449ekLkxbLOxOyHW2b97oaHM9GoSMTq4whA9v5c8S5fAsPwxPWbBiIP63fK4455rhZS650OKeeemrqbBgew95nhThL8efNm5e0EDQRNA2eO3N5Ob4OjRNXJh7Ih87P2TSQZ5zwDmFpNBziE3+Ox7SS58Lkc7XDrsaVIt2ctnirLv/n3yEO8XHSynHmc85bNX/Znzw6MumSb3g58jt4blsA2wEcd9xxcdlll8UZZ5yR0s7pV/M3E6+bW4b2r9IOEZ/FixclTdH//M//xLve9a54znOeE8973vPSPlm2XLA60LvDDUm1p5b6sHHjxkRI3bPiTz3hLzskyDPaptbm1nx71LNyjFC3RvbGj4EDgmUl5Gte85p4y1veEsccc0yqlyOH3HlPCsHaeViXlAoCuyUCOrmnP+NZ8YyPfyiedmx3HHng+ITsVIHV1z8Y/3Ppqjj++Gdt6ZSnKu7pFI8O55JLLklk6t3vfnfSXBnZIwemT2hkEK00NbTZbgZJES4TEf8zWcqEA2EyVYR0IEzOQx3gdHr7yeVlrPe5z33uE5/85CfjkEMOmTad9+TeOKKleYhoi4fl04L582P16nsSEUfUTz755PjXf/3XRKjue9Qxscfe+8SBixekVYGL5s+L+fPmpbq0xx57BHzUk0xUM/lFtBAxBEu92hFOfXzGM54R973vfeMd73hHPPe5z40TTzxxh6U3kXcoBGsiaBW/BYGCwHYhQAg/64QXxClnvTfOf9NBceCSkXdc364ERghkHL2+sz8uvGptnHnGIVHfsHl+cgT/M/22Ts40yWGHHZY6PBqYxz72sdu8FjI1mqslG9X/1evR4phtz3TitjZg3D1b3PwFC7domiiM5s2fFyvuXpFe7+CDD47vf//78e1vfyde8cpXxu1Lb4+5z3tDLD72r+JPt9wWDz2oPf5qj+ZEykyp2hX+QQ96UCJniNRwdWy4e1OFpbiPOOKI+NznPhf/9m//Fj/84Q/jox/9aBpkZLI3VWlNJJ5CsCaCVvFbECgIbBcCOua/eerT4mNnfyTO+u6yeMc/7L9NPHlGgbapu3cwNnX1x8bugdjU3R89fYNpBWB//2D0OPoGo7cvondgMLq6B6N/0JRWhLCbzT3S2T32H9fd1hl/WdcQxz74Ecl2Y5uEZ9mfdevWpSkzHU7u0HZlBzNb4IUlgjWbCGYii2kqbmi+fl5He2zYuCGRLnVmbntHHPnkZ8WDXrMirvroW+NP3/1i3LN6TSz99Fvix4zb6yNaTDEPDk0d2uD34Q9/eBx44IFp6g7Zd0iH5jNrtzKGvf1bNyRNbZWt1eZVhKYZbUKa/fYPsMOqS8bwpiU1+J7enujt6Y2enu4t05Rr166NX/ziF2kLhz/96U9x4YUXJu3trqqHhWDtKuRLugWB3QyBL37xi2mPnEuv743fvX9FNDfnkW4DI5CIuoZkw1pf3xjNTa3RwFamuS19YsNybp/baG1ujDltTdE+f3HMaWmIlta2aGtpjqbG+mhpbY+6/k1R1zI/Ghqbo36gJ/p7O6PrpmXxlBceFm17H5dWMM1W2LN9zN57772FXM3Wd93Z76WjRxByh7+z098R6bFfGugfIkeIjc/VrNswZAvJXq27byCuuacvuo97RjQu+Eg0Lto7Gvc8IDoe+tSYe8RDoqm+Pl58aEfs1xbRuWlTmqL7q7/6q2SPZXGFBQKmB8XFDssUNRut+oaGWNfVG/9z5R9i5R9/F/Me9LgYtCKxvz8akhhoiujvi/6Na2PFJV+Pxcc/LRrm7RFzGuvTzu71aUVif7RGfwz09kR/T3cMdm+Kht6uaOrZkGwQaW0RMfZiu9IVgrUr0S9pFwR2IwTYc7zyFaek3dob0rLrxmQT1dDYEI2bNS6MbY2erTgyLdPQaBn5Zpsfo+2s6soj72Hx23Ya8KRh/cy+mzZW1aE84AEPSNjNvjfcdW+UCdauy8HUp0wrh/xk19TYFN09m5IG6+KLL46DDj0iblrXmghR674HReOivaJ1v/vGAaf8Rw4St639few1sCpe+cpXbiH1doRnD3XdddfF//3f/8W//Mu/bPHvwrT98g3d8c3ln4xNV18R+zzmedE4f49t/PjTs/Ku6Lr4a9HVsST2efart3mu+WvljXV1wbKrpaEuOprq4/7NXbHi1OfEN775zfR8V0/pFoK1TbGVPwWBgsCOQoBmxTEpNyqxmlTMMzow7dXdd9+dyOmhhx46qzQt06FgEKwdZaS9q94vESzTe5sz4L+pOe4Pf/hD3HDjjfHyk18Vt6zrjVsXLImmeYvupQHuisa0DxViL3zV+c6h74fWOsRoXnN9HLxkYdzQ3xuDA8NrmRrmzoumRXtFY/uC2ijSOCuF2jzgYi6woXcg7ljZGeuiJQYamqKjZccY1d8rM6Pc2BaRUTyWRwWBgkBBoCAwfREwDcMO5sgjj5y+mRwhZzQpCKLPIn35y1+OK6+8Mm644YZYvnx5mlYaIdhOvT3bbNkQojTluZlh2XHdpp8cO6qrrrwyDpnXFGc+fEkctmReNLfNHdpbZTPqzQ118fgj7xNdnUNar9rCsJeYxRbDueaG+njeA/YNmuy6EfZisOFoyz4HRUPL+BYWpOlFG542tcRfNtkqYriUd+69osHauXhP29SMgAm573znO2nfEw2MenUkoZLtPfLz6lx3vs4jmqyG9t8zo6R8zs/c00E4a/Tmz107clrA419Y9/hz5odLwmIzwtKSN2dTTXkJsWsGlwxW2VS4zvnM4Yc7u5ePalrZ73D3qv7z8+H8b85yORUEJoWANuND12NNi+Q2pN47qnVSBrRD33ljJKzdaDM+m4K4Oef2sr2ZzW329ttvT3tNsQWyG/c//MM/pLS8w29/+9uk/VixYkXKw1Oe8pT4p3/6pxHlkbyQA/JuRZt2bSUcrVP1/aTNLsjeTdnw2jtW/Yz0XnCYTS6Vf52NpjCRumiodz00vW716apVq6K3pycO7GiJh+y3IJoP7oibO++I29dsjLp5e8Zx+8yLhyxpiys3bgzG5VnD55M4onHP9g3OtU4dnN80NOU/0N0V/ZvWD6Wckh/KA+P5prnzY7C3K9L+pJsjQZyG4071m/NuR/g2q4U3R1Ob9s78P7tqzM5EbpalRfD8/Oc/j1e84hVJMJlHzwJIQ8xLbwkigiYL5nxGZsTBucdPJk/5mf/u54bov2fuidchbI4z3xNnfpavU0I19/M9Z3nJRyZk0nPtyNf2Z3HNuS9tTnqc/DnyO7knXmHcy2k4cznP2X9+N/+9t/+uxcllXN13IH06yHzwq9Mwuv/rv/7rRAxTwBn8Y2T7s5/9LJ785CcnvJSBznE0Qj+e183lomwy7uMJNxv8eHfL5ZGgXLdq3wsB+/3vf58GUTRDyNJrX/va2H//rSs61TUbN37605+O/fbbLx7xiEekFWHqKaPlJzzhCSmdO+64I9VJAzGdcW7TtWkO918+zzrrrLTxpLS/8Y1vpA90P+5xj4s3vOENYUsP+VAvlONtt92WCJ/d1OVJp61deOfcZq2epP266qqr0vsgUD71wjYIgeP4Z6f2/Oc/P12rb/w85CEPSeQO0RrNjfV8tLDT8VmSbY1kUnMiMM2VfbHgDFtl5b1bm5tizW9/Gr+++OJoam6JvfbeJ5bOnxdv6euJpbffHq999Wuid313+KYgw/m2eXPitmV3ROu8+fG1i38Si9qakp1UQ31dWinIqH7lylXR0bs+DrvqvGhqnZP4UF0MRl/UR1f/QHRt6owjjzgg5i5qigccOn8LX7p1fW+s6uqPJqsKBwfTuWdgMD2vn9MaPXObYklb4xb/uxL7QrB2JfrTKG3Cx6aEdnu22uuAAw7YQjIy2ZDdfJ3P1Xv5darP8r18zs/yeaz7+flUnb0n51w9qvfydfK42W++Vw1TG1f2Uw2HODisnnEI46wD0eHpRPLnOdzTUegEjB6RKp2aEfnKlSvjW9/6Vhx77LE5+hl79k4MXz/72c8mben73ve++N3vfhdf+cpXUkc9Ut2ovnAuB2TKtc4AljpYuNkkUV3WQe8uTr0yKKrFL78/e5i//du/TR0mQvSXv/wlDj/88PQdt+wHdt/73veSZomfD3/4w4n0ew5j03f//u//HlaLISY//vGP48wzz4zjjz8+RcEP7de+++6bCJy8KB/11wanHNucz3zmM+lAzuzbZcdwG0RmovahD30okS67iqvz9lgy3fTEJz4xkTMDEfHSuCOV9j+Srg1VX//616d3fOELX5gIoq0D1JM777wzTTm6Riwf9ahHpf8XXHBBereUuRF+pAXb2eR6N3VHU39TLIo9Y/HBe0c0tMdNt9yZXhEpVRZkFIeMPvKRj4z3nfGe9D8PYMir9/zXu+MlT/+HuPvGO9Iz6qXF990nLrr6kmg6/MFxacv944g9W+MAxGdOQyxqaYjDFzTH8ltuiltvvSU+eea709T2UOCtA1j/P3fuuUlWnnr0UN3xWR27O/Qm+b1Vk9U3QJ5H9HdtjBcvaIvGqsorR7wLzruP9NkF4M60JBcsWJAqOu1VHvXNtHcYK7+588nnsfxPh+cI2Jve9KakWZwO+ZlsHhAfAvvRj350qmeIgdG0pdXKxTUNRe6cpZfvudZBEvw6c+TVtZG2eDzTMdCKIAMGCruLq+I13DvfddddiXAisg984APjC1/4QvqsSNWvRQj2DuLnU5/6VCL/PhnDMXx+//vfnwYB5557biLHPn6M1NJ0cbfeems87GEPS7uen3TSSakMM6FBqpSjOOyKbtqv6jzLzpJ+dUQZZ7dkyZJEoJAdTrxWqiHWOvonPelJaZNJBMumk2RYDq+eaEO2D3D23Tok0c72b33rW3MSI56lidTNJte7qSdW37Uy1ty8InpXdsaawdUJsywbvS+MOeRSu9J283P3YR/dA3HPLcujuXGrUXlj/dAK4Nt6mmNlfUP8312IWm/6PFZLfV38zQFz4u86hvbHoiEbTjsI8z0WLQra0q18iXyIaB5BP7W+ty4aK/VoV5dXIVi7ugSmUfoEkoqeR5HTKGu7dVaUx2wS7oT14x//+LSUW0etUzalg1S9973vjWc+85nJlsg7X3vttelTL3/zN3+TOkN+hLFTM0HPLw3Y0572tDQdZAXdmjVrksYkE4PdpfKMRbBogv7u7/4uaaBM0b30pS9NhKeKD8xND8FRvctkhh+ki7aI5ss0nmnGz3/+82laL6ctDO0GzZSyyg5x01lfc8016RY7q9EcWSSOqgbSN+bIp2oHz55Kftlbma6UPlL1sY99LN2n+eK8B3IlDp+7Qa5oZJAr2k5kbjQnPLvN2eTa5s+Nwfq6GBgcjO5NXbHszqUxZ97QdCqMfUqnt6snBvr606eSDGDgUMUfHhtXr4/+nt6orxCstXetjP5N3fHw/RfEtavrottuwKkcIvqYaDE12VxXauPLGLtPVmSSl+/PpPPWFjCTcl3yOuUIqMyEFwE7UoWf8kRLhONCgFDTWVU7rHEFnKaedJq+HcaI2jvpzHWUpkDtm/OqV70qXvCCF6Spnq9+9aups/zHf/zHpHGhdTFtZZqJzeDb3va2RM7Y3/g4rec0sKa+Zgte4y1G9WQ0p7N6z3vek+yNYPWjH/0oTb8MFw5ZUU40P9khVMgJTdONN94YyuTNb35z0ljxQ27QGJqmo1Wsxpunmn75y18mTZO8DOeEyRrNnFaOm4bKdHnOE780XTTvCLr0pfP3f//3SRPju3SZOHlmhSUyxs7LtKDw//Ef/5HqYTWvI+UrxzXc85l4r6mlORo3b2Wg6mzo3JimcWHlYPR+3Q9/FTdc8pv0kWZa4lqHfDlqHTy7u7qDvRQ7qaprbaiLx+47J03jjUWeqprrahyjXVdJ+Wj+dsazosHaGSjPkDRUzNlIsDRio1qj24k4mpB3vvOdae8mHQcBC588itb4TWvt6I6csJJ3Qm82OHjRJGQHR9OgNFOwPeGEE5JN1g9+8INEAHSoViKdc845CWvlqbPUSZs+yJ2j8nr7298eX/rSl5IG6//9v/+XyNuOLp/8Hrv6DIfR6ggc1F+2SeynkFEEliE47Va1fajvtEhVrYX/ykl5vPjFL07kyhSb/9IlP5zPOOOMVFZHH310Ij/aHkN1z6QhTveG6wi9g3rw9a9/PWnaqmVHG4UQWX1oI0sG++oOY31bOsgLo3n1g7YJ6eKkKR7TmKYU1Y8cL1tHmrLRcBOHfA2X311d5pNJv6HJlxM2f2cqBmPDxo1x6EJ7XQ3JGZv90mCt+POd0bViXaoPnZs645Zbb0ntlO1cV3d39Pb3Rbcd1QcGtkzc1Q00RF1jXTxiz+a4bV3E7+7pTp+0EvMB7S1xUNtg9HfHljoz0nsoJ3VlIi7nfyJhdpTfQrB2FLIzMN4sAIfLeu7EPNsRFZhQlAaBWXXuMf62bNtoleC///3vnxp41d9I18L+9Kc/TXYldhI3VVGbhrCEft4UL081GA3TlHhGYOv83SPQkS2jYca94mOfoqOhOZnqkS4MCJrcKYz0rjPlPoEJL+/j3dQnKwrtHu1DrYjRq1/96iTQv/vd76YyUPY6Vi535MpJpysuCzTExSaEwby4GV9bDTdbcBurfL1nbkfD+YWP5xytoekxU60vf/nLE07ZjspzfjkLEqza40zD0TIiLAguOypThMgOmyftUqf70Ic+NJWZsjzxxBOTtovtDnJHc4YEI1DKjrZRu9KGMsFD/milcrrSVkeUN2P4s88+OxnrM2y3WEJeTP+xHZOm+sFQH3m0StCqSNPFpkdpt5A95I8/6YzXzbZ65H0G6xLFSgbi6zeuj/aO9i3yvaGuPno2dcVAe0usX7Y61q1dF9ctvDrOOufjsXLNqqhraggk7c6ld8aXf/rtaG1sSdqswYGBaG5qjltWLI2716yKBXssiQM39sea7oEYGOiP9oWt8YWrW2PxHkNkbrT+ZHtI7faEGRUU+HwAACAASURBVG8dmKi/WUOwdG6Eh0aqwPKoSiUarQDHAoygycJmtjWw2ncn5EaqnHmFFuGZNTi14Yf7n/EbCzsGuKZ8CFDlhdQ4hGdPcf311yfhaARru4KXvOQlwxKl2jwYsdpfx8j7N7/5TRq9IlnV91RX2Jf813/9V4qTAazpCKueLrroopQH2MgXw1yGve9617tSHO6ZannZy16W6h+Bf+qpp24Tf22eJvofBjn9iYadjv5pmkwD0jogrDRRcNQJW1n4zW9+M5BhnfYnPvGJNOX00Y9+NP77v/871QkETRzseUwZMapmg6XDV8/IAZ2uzjy33emIw1TnSR2B50gOLlb9mepDWGmAfvWrX6W2URsGxsqkip+6/8EPfjCRFmRG+zSAoSnSti677LL4yU9+ksibaVwDFlO3ppb8JzcMkLQv7YqxOQ2SdoaYIcNI2HHHHRe0l/xXnbI97bTT4tJLL01tThv0WSDvJYw4GeazD2Ngr8171/b29lQvTDki3W984xtTXUHgvGPuM6pp1V7DYSwZVhtmuv+nTdo6eTcY6zds2IZw9vUO9aneo6mhMe5evjLW37wiXvbkk6JvsD8a21pizr7z43+/f0G8+d/enHDs7+1PH2QeGByIcz57Tjz3hOemFaW5LuX6pH6x5bMKNN8bDi/PauvBcP6q9zJRr97bVdezhmBp2BqoEZWGQDATstTh1c50okATIDp3cTCMVFFmsyOkax2hbVmz7QPgqyPLlV4DGAkTzwg5HakpCOUykl9ai3/+539OnaVRJZJCyOp0n/3sZ8fHP/7xdE1j9O53vzvlZTyjT8LbOxGshDvbHkL66U9/erovjzoZz21PQRh7VyTsqKOOSmlmPNgJsSFhcO09xIt4mtIwlUh7haSZ4qqOvnP4yZylNRJ2k4l3V4RFaNUp0zlsYZSJjt/+RMirFWE0hsrlqU99aqoXcGXvwx8C9v3vfz/FYam9MqHhpJ2wOo7Wi0bLNKQOu7ghBOCO0CA8ZKOBCvmYDcSrONH4GPBUHa2XDpFMpKV1iIdDXmyzoOPUJjxDkGmrap12q8wdNGQ6WmVeJTCmI4dzyLj258hO21A/bETK/osjr6VTKyOsXD399NPTYgp55R9pFMdMcqPJ3XG/BxnWvPW9++sHY/Eei7fImX5Tv5sja2kaIrtJsTkY0RgNEZ19seL6pdHS0JxIcu4TBFHXWttakxwdbrGJ/DtGI7ee296jtgzHer9qPsbyu6OfzxqCZQkvTYIz4WyE5pow1ng0OAXmnDvHfHbfKIjQRwh0pA6jHyNlAoMq2/Lf17zmNSmOHV0wuyr+WkEDGyu5jEbPO++8ZANh1Vf+BIIl2va8QV5qHdxomqjrdZxGsTrA4UYYOkICVuNAlmm0dLDK0H35Ul4EOuIjD3mFUG261f/Cm8pTfrQcVi+dcsopiUjZNFG5E84MdsUnLYbT8mo0XG2srgl4Hbl3UF/s12PaxLSDe+qLfYKQrqnq3JWBY7YQLJ0v4ksI69xoQ3S0OjtaDW3UoIYhNcJrSsjGpJb3mz5SlxBwGosXvehFCXf7Jen4aR5ptuyJZEpRWjQau4NTl2E3klMfkVj+cl1yztcjhcv3tQ2DTHVxJKedjkSOhguj89Smx5uH4eJwj8xHAMeKxzuQAYg5GQML98bjhpNb4wm3I/yQf+QrclyVURNJK5W9jZSbGsPu6+vXros9l+y5BcO+ivE6e6wBO7RXdF7Sqq+rDx9rH85p36NhBvdNGzfE0tv+HPvtf2A0NjSKMJ3lzdahvb09abDa6XM8ph5bWlJ5pTooO2nD9qFd2+vqhkwO1IXp4kZujdMlh+PMB42BymaKyWZ6NCIautEaOxqrlmhDFAzDWIIAs1YB3CO8NXb2NUZH5u2tQjHKVoERDGHG2xjHme1p5S2TmGqmCCAqeYawOkB2DEgrWwcdGnU7vP0fzikDgp1GwbSQJdEIDqJSdTDmz1SBES0tkI6xlqQYdSIxeeRcjWO4a++k07388stTnMLZCoD9CTsMGhGHaY4//vGPyVCWoJYfBM/0U7XTEl6dIQDEzdhXPaExVX9MWSGV0rMVwVTUF/UzCZThXnAG3lOmyh+BohVVfwxoaECQcVNPOg9tV71hhGzFoaleuKuTHExosmDs2jODLFoadUQ9Q8TUu93FjVZP4OOYTJ3McUwVnlMVn3jG67RbB5uuXG/GE7YWN/WT/DO4QvrJBnXbkeWZek4G6vTztf952lK6GYN8lk6+zu9VPbs2KGQ/Z5ENEwppZz/jeZfsp3+gP+p8xM8Hl5vqUz7FA5fevt6o36zZa2psiv6BrStKc/iW9rbwDcNaJzxZPVqefPdw06aNcdMNv4u/3HlL+pC0UsxpYk8HH7AkBvo2xJU/+3FKgiy287zy4+o2Y+V7iHvtvV+0z180KqmrzeeO/j9rCJYKzeaCQNZx6iRNGxgd0zBQNRLgCsh/Whn+7QVjCssImkpchTA6QB5oMnSWRsn2TMmCfUcXyq6MX8XNDcz7GoXQHiBGnlmpo3GbzkM0dIQ6M36zANK4vv3tbyf7iGxPQWNBq8OOhkZCeIImO0KJkBAHzZKl1J7TOhFGyk28Rp1IDI3UeJxwjHCRKHnlkDhTjeqBOKWLKCF0yDdCiSAh5oRYnnaABQIur5y8ekZw+paa+mRq0TSEFVo+/TFVTj5ni1O/lKkOSlmoPwZIyBZtaBbKzspPvVPXaKMcwuqwtHn/c4elHPhl92PKVx3Uie4uTn3UqRU3PgTUr1zXxgqh/WX5lv2SB+SFQzzqXyZWZIS6q94qE2H5ISfFlc+u+cvPc56E46p++ckHfwYiBrts6WxvQvZMxMn/suXLoqGlMfr7BsK2DeLNrr65IfY9/D7RvWz90K3Ks+ynobkxkZz8P5/le6y6WBdDRI6/ns2yOYevPff19mx7S9ltvpMlI21Y29yhxRTbet51/2YNwVJJqX0ZUCpcB02FjtUI2GiZloVQV7kVqo7UiNn0FpJFUNNk6FhVZNOFNuSzIzQi5ltYKrLKX62Iu674pjZlmHlvzruzSdKRwcZ7m24jRHRqVuax4WAjY0qGfxrC7GCVy0QHqgxoAxmmMmC2YswmkRlH6SC1tFfsORAxRE4ekCH2TTpPxuhWPcnHeJz4GcIi2x/4wAfSZ1RoqnwSBMlDrNh4IUqEIgyy4FQPqiRQuWdNVU7bO9qhGink+BGvz3Lkd8t+J3MW11TGN5m8TDasOmaKWVtDlGmMaZiVb9Vtz/tq16aElUMmYdU4Z/u1+lvcjkGgtj6qX+QRolNLkuQgt9naMsn/hzu7l8mXOPwfzh/zBDZnNPG1K3LH+/byffvSpXH7ijvjgPa9oqVt66KClI8YjPsff1RsWLo6lq64KzZ1dcaK1SvTFF59fUP67uCCxUuis6szbaNi+i5jxH7LgNZBnsKn1vHr24XD8LZar/f+D5eau/39fXHH7bdM2GarJpop/Xvvt57S6HduZCo8wcrpgGlKGEczyqZKZTtjeTCyoNCz8Xq21RCG1gGh0AmrFAS2JcA26NMBIGxW0+SKtHPfcMempuPL7wUfhIdmDy7U4FYKOV955ZXpnqk3miZG4UgTexp+CQidJ0Jm6TZypNODp/JBypSNTjWnR9PFvg3xsbSaZkN8NI/wlw95cg8hzkRwPIggfpagI3nSRxBN50lHmVa1YfLDj+lmR9VJUz2qOu9kxVJ2/NB2FjcyAjBGsLL9HvJLUzmcEB45lpGfKN/d0Wl3tSR1d8RhR71zLbbaOnnn2BXO4o5sBK5NZVk63rxoJ/vut290NvbFIQ8/MuZft3XLCuSLa53TFgsfuDCO6F4THzr/k/HDqy5JhI9NlmnDxjmtsXbjhvjpVb+Khs12tPW+29jUEEtvvy2uvvrqJLPZ3tbmr2Gz5i4ZUo0306P4U//v+svy2HuvA++V1ijBduijWUOwaD5omapOB8rg2AdCaU5osNjLMFanWrUihjEsZ0SAbdOs6CBpTRAJy/w5lcM0l/+IxUQ6+BTBDPnJjYBGh/2KKTJkCYHK5JW2Ct46RNNrpmMQUAQ2r84Tj4UGbBQYjyNHNE8wFkY55LRAY+rNKjEu38/kB1HLLj/L/8c6i4MGy1F1E42nGnZnX+dR7c5Od0elp+0gyTSGOghaZfXD9MpknXLVccyk8p3sO+fw3tmAsLgdg8B0q1NMEfI+ZFVN+3jf3sDTilvtBaFqq5hskNPaKZnPRstndV508ovipOecGOs2rI8NmzZS18cXbl4XX/vSebH+r18YDa1DxuW+G/iUg+fELRd8Op79tKekafrhsKNxsp3DVDlyctnyu+OQQ4+aqignHc+sIViM0FesWJEqCpUprYmVRVYlseugvWBbxeDVpoYMAy0hRiAs+2e7xQYEC8a62eDQVqjEbLM8t4LG9M9UjbQnXXpTHIEKWm0IyAmNnqlXnWF+hgyxg8n/aYBoluy2ze5F40Ro+RvODknDFXcOn1+j9v9Y9/Pzsc4jxTtWuPJ8xyCg/LVPbY39CsfInQZ6KtzuqsEilxDV4nYMAurtdHLkqxXVk3GZXNVqPw2izcHldyZDW+a0xh4HLIm9GvdN03MWFT6wbmlc0DY3mhbts823CH+1cTD2mduRlBUG1MM5ad5rnm84j+O8Z5XjipWronmEVY3jjGZKvc0agmVFYDaYtcSe1sT8uL1waKyMlBm8ImK0Vab5aLfYASFRbLQuueSSZDBIW2P6kKGy/Ztovmi1VDZhkQ0EYba5WiLi/3AjI/erfuECE/gzhvcsE6is9ZptWO3O74OIO7LwnSgW6gdtMEJg/yPTx6bkh9svZ6Jx87+7arC8e+q0xglaLkdhXM/Gtuq98vtNdmC8vfV9nMUxYW9VGTzhwJsDmGWwHxl72mp896xYFT1rO+PuP94Zex26/5BvxlJkP1MSi3wi4rAFLUNars3Pcj7u2Ngbg5u2Gvfn+1vPg2Ej08bGew+0t/qZ2BVt2O233xHzJ7A7/8RSmLjvWUOwCGwNQGOyVJsWiybFqjdbMKg81OcOJMm0F4JF42IOXePTGI2i7SbNP6NsK+gcwojzda97XZreohGbbc47VxvZRN9P2NkopCeKw2z2byqBDSMjfoMPthWm9iba+agrVj3ZUsEKQitDp8J2TbzI/kx1NAe0ULQTEx3EkV/COo/myEiaQwt8tFfXbBLJNmEnIwNGS3e0Z9KVLwdHHo+UD368p4M/7zCSX32AveospFBfJ+NGSmMyce7qsAYj8GOyUbUlXbtmbbTVN8ctV/4hbcOQy6c2v/fpaIyWyuxGfj4wOBib+reae+T7+ayKInfSrjenOAVOvVi56p5obd062zIF0U4qillDsDQ0q9Voqlxzzla7ccDPDYTgYmOUl4qrZKa53DMv7czux8GPSsDg2xQkWyNkrriCwM5GIHcsyIy6nOvzzsyHgYnPAZkyNxihsbRa1yBlIiRL3rU74WiWLTSZqs49Gx2LbyY5gz8LRiy7t3o2L8rI5T3Wu/DHnm00Jw3k2N5j9hazdY3d3O33xuUd1cnOav1CWnP5qoeOrKUeKb1qebqWtjhyPMK5j7SbIrboQT2Qtq1xshzP8ec4mCJYvITo22uOqYf8yZM0qoRLfAbMFlEUd28EzFDYUJkZCLta+CnXVXevjD3nLYruTV1x85V/iNZDF6XNSG1IasPP7Ho3ro+9F3REZ2NDDGxeDthQF3HYvJaYM39koqNuHXTQgfHKU4a2/8nxTebc2dkVPb290dY6eVvOyeSjGnbWECyVgmG6lRVsrkwPZq2WF87CoraBV8EY6ZoAOPLII+MXv/jFsJtfjhRupt2vCsSZlvfZnl9lYxqNNpWmx1YWhKN6nev2zsCAZsU+Uzb0RbYQAVooH97NxGY8+ZBn2mULSexZx9iWcK/tVMcTV62fbCw/UVxgrJMeLtxw96rpCjtZJ217+Clj9p5sRe3Dx4wBaRjNyd94sENk2KEaRJKFvmDAXIKTPi2WQamNM+0Lx4+Olz0rXPlB0JAb5IyMHQ4b/uxZZ6GROKwCRoq8S3VFmfxYCGNxkR3WH/7whyfSlTt6uDqkob6R67bXseecjWf9N+i1XQH5rB6yrzUw5ph6GChrL5N1w73nZOPc1eG9k/7S6niLt2zQrd7ZH2vJoiUYcHSt2xhLL78z1qxcEcv/eEcM9PQN7X1VVxe//vUv40kH7xl7HrEgegYiGurr4kF7tMa+zX1x7hVjTP8NDsa8eR3B2H0q3Pr1G2JOa+u00mDPGoKlEWtERmWmHuxhRC08HqEzVuGqhCMJkrHCzqTns1GAzCT8R8srko/M6KRofnQgDloko3edn07YQCOfxadduOecy3ekczV9foTjxO2ohmPbaK8zU1o2bCWU7Sk2XicuBrq0wtLRieYpn/HGMZw/8WYyMlHSY8GLFbEM8GmPbNMBazJEHsXtgGV20kAm8uG/aSlkwBk+ys67MTPIeUIg8rU4q/HAEqFBYmizEOrxbCIpjzTtOd6cx+oZCaaBtC2KBSgW9mQnT6YKzQIoS4t8bM1i+xTvx9lQF7n20XemGLRDSDcnXe/CeXfpMM9gY+dj3RbCCJ8/6swfDRMzDZo7W77ksoMxbdoZZ5yRduungULoLEiicXvWs56V0oGND8DD1sKkr371q4kYIgvIm73vbDSc85gCbceP98/tYTuCT+sgsEaykPv8hQv7XR176OFb8t21flOsuHVZ3HjZb6KxYn983TXXxn5HHxwvPmz+0I7wlBlRl3Zo7+7q3KatbIls80Wjr6hMoZW7ur/vfkNb79Smtav+zxqCpfKb5iMYGVsbBRJUU0WwdGajCa5dVYDDpUujQCNgn6k8Uh3OX+09DS0L0tpnu/N/HY+90dQl+3pVO9idgYt6p34b6VvF6jp3+vKSO/7cueVzbd5Gul/rb7j/ue47O3RotjqxWlTHZ3uOiTj5prnSESMfVhqJcyJasJHSy530SM9Hug9TU5/swZACHyvW8eickVj4iTvj6Ox+PpBQ1zQ/tHPeBUHL/nO61f/DXatvytmiG2XOVmU8BEv+aYpGc/yYVpOuz16RFbnN+zQUjSJZp7NCjpA8e9Z5L+VjaxtaNVuueG4PM2TLdjc2B3aP49/nrs4///z0cXVaJ+TILIM8ZEdWk1HkFe1ZlcRYZHTOOeek+GlN2cB6ru4grdokAmZDYljnD6wj7T6hZCsQGjPvUsU5pz2R82wmWHBQJvCy0t6H62nHj3jwA2Pl9Uu39nvavinCuq3a2vUb1kf/xp5wq2EzuRZfW2tr+hKKNjGc47WpqTlpyIZ7vj33aLAO2G8fSrdp47bW9GmTpe3LiIangZlG0bhttUBQTZVjmzXZRjpVeRkrHkLTqM6o0cddCRmjOLiM9g4aGcGVO9Ox0pltz703QeqAUz7Uo2984xvxu9/9Lu3qT8C75+Bnezv0ieAnTXVwqlbaTSTtWr9wMm3ko9z24aHpmCgxghsSYwpJnVM3Ea2pcOLeHmeKCmHwfvAmU3IdyPENF3f1Xvaf7+VzDj+es/qHcJqKo1WoboQ7Wnj5pTEbrf3m/InH/oA+mK0ee1/af5pJ8gNxYQd1wQUXpPTNCNBYageIj/IyTYdQyaOpJWSKPZ58OCwS4pcmy0e6aZjIo6pDftQd04nnnXdeem+kVHikSkePNIrDSm/XyBxCT977nJm4yS7EUFuUbyvDXYsTiZ+MgyeNpHzNVgfv008/Pb3ny172sjj4oINjn8MPjJ5VG2Ld3WvSqsEh4rIte9nYtTH2OmjfLdor+PDhe4Jz5s4dCrcZtGqrHBwkX7dqgqcC102dXbHnnoujfjvb/1TkoTaOWUOwCA7GstTKpgqdHRroZJ24Nd5d5bLAzGf5qF7nfOV7BJsVktTuRqVGgRqNPcEIouGcdySQCLypdDlP4p+I0+kS+jm/mdBUyU8W5KPFK31hONfidOT7OnXv7Ux7YdqDobCpEoSG8HfQhtr+gz/p0nL4IDjjUNi6V31H8Vf/5/TlRfq1z8Z6B/4nEma0+Cb7TD4QI52n8vE+2+PEQ5tC40CLoWN3nqzb3vyoB45d6WiJfLkAcVHH2Bipi+Nx6qApP3VvPI6cNI2WCRbyjuiQmYgW+yxT0MpJXpBqJMyH32mbfB1B2dEkkRvaDU2U6TlxyM/y5ctT/bCJMK2cMw0JbYm6Ywsc04bKzCfJmGLQGCoHX3VgwI5M0XSZvqX1lA/vqL16Xxo+9cYhHNmhzYoTnpOdHpSWaVN5m60OVto0ZwpW2+6Nvjjyrx8ct1x1Y6zuXR+DdRH1jQ1hB/bslq+7JzbNXxR/WNsT7Y3kWsS6noFErFoa6sICwY19g2lLh/nNDdEzMBDd/RHzm+ujxTR5jmiSZ3tgbdywMY4++shoaNy1bbj6KluRqt6dgdeEQLZHMdowOkawxqNaH+t1xT0eoqbzzERAnMKN5cbyo3ETgo5sz0FoGJURJK6lyV9OGwnw31QFrUvWEGSiMVKevCO/wg7nchriIciGy3tOwzP5IXxNsxB+Vf/iyum4X30mDlNPRsZGvPzZ7NVoWpnySxgYgdOeeO5ebccKGx0GQQwv/00hW7AgDdMWRt1vectbkpD3bS/+CG1LutnwIVkENJsVHZB91MSBbFnm7mPiiL286DDkQzo0qUbY6qR8yaPOCOFlVKzzGa/L71fFaLxhd5Q/eaHFmIjzHlz1Pdg5XXbZZakcTQGZWpusq60Hk41vZ4an5VEfDY5MqRnYVfEaLS/apJXUGefR/Grn2hRipC3ATJ3PZaBua7Om+dR308Dyoa77Pqvvb9J+MUMwW+ATWtqHgZ1pRO2DtilrzT2jTfrBD36Qphh9M5YhvzRsEaBtm9bLA1n5YgemDQrrnWiu1BeaLJo0Gj42YKYIaemRQunBITt5JjMm4+CDxI2nD5hMOtMlrD2xyE0y2CzI4Y9/ULQdvCCuvPk3cfgTj43WttZkZ9Xb0xcrvtITH76xK9q6l8WcRoSqLjb0biVY3qmzfzARrTkN9dE7MBj9g4OJYD1mYX8cMTA1H8oZGOiPtevWxR6LFsXcubtOGVJbhrOGYHkxHaGKoVHpnCfbsKpg6SjHElxGcASRTpRQMvLMI3x5IsScHdm5zp0tv/laWq4JDKppc9meG5k5EyLic/CXr8UrfZ2W1TqEGOJAq1VNN6dfPUuH4EXQhnOIgxU6VvwwuEbgqnESRFZxEnwEnvl83x002rXZq3xz3g2JsSzbaBRm8M3PGL6y57ABHjsR+aF9I3wZuxK4sDbSV97yYlrAO2YnDSNmeZXWK17xiiQghWPEbCRuhG61lDipx33ImeG1JezyL3wW1gQ+LK+99tq0KsnInL2KVUuMdHUq7GUQP+H+93//NxniMko3+uaUEQNfI/xczjm/o529f+6oRvM33Z8pB4MEeHgfDoE1/URDqCx3d0ezw9A4f3JqInjAVD0dS055TltILhgEGAhpx8qF8bhvtyJW2rf6jShZeYh0aXNW5TEeN42IpElX2Rns0OrSOpnSQ0iqhJl8MXVHLmjzBn/su8gBxKsqS8g5H2FHNp3FT37QnsEGMZNf4ckQAx15/dKXvpS0ytoaRwaSWdp/rnMTwZRfeGnvbPKqeZxoPDPFvzphwQGZaEWpD9nPX7ggVqy/J+btO6RhTLhExNoN62Ogvin6ewdiQ++2b7i+9n9s3d5hfU9/3Ni/Ie43OBjNW7vDbSOYwD9lpP9paGyI6kenJxDFDvE66wgWYZE1EFTMeXQ2GfQ0qvGMJDVgjd8oz07yCJ9wiBZSQMDoUBzjbai1/mr/5/fK9wkqH7SmqkckMgHLz7P/2rPnBKLOHJEazllRhDgge4xffXuQoEboaHCQERhYVYa06EyVh5U+WeDleGl4LLPWIbAx0cHmPNLWEaZG1DkcgSt9NheMaj3XqKRttMz43Oqm7BBF2ibCgX0IMmVKBLERzpJu5A8B0nHoFGizED3l5ZzzI07h5EdnoCMxXSDfOhR1zPtYLk4oEfbyLTwyRegTzurCeDrA/A75jBRKv5qf/Gy6n5U/jYMpL9oZdUc9OfHEE9O3Q5UBrYR35G88TvnZC8kKu2uuuSa1M3Y4Olpp0P4ok5no2Eo6cr2fyDtoI9qv9qOujeTUI5p99ZLNmXCcNLUvbYkWwywAHOGtvam72re2oT4iUWSAwRytrEEl0kM7RU7YXwm5sZEpMwVlRdMlDoQnD6pohw1SpFF12jOjeNNV8kib4qzdVduCd2F4T+PlfjUe+3tleVGNeyLXMEDwbCGxuzj1x8D41a9+ddLe68vUD9/izTj06yt6umNO25ztmuobrKPZT8sOJw2rMiJrfIC6TBFOGs7hI9BwNSgkA8kxBaTj3h5hVZuChk3QjOY0bKNCBS3NLASc8yF8vj9aXNv7jF0EWwaalKqgGU98Oj/5riVY3hthsQSazRFjYKTF9B0yZTSL6Og4pUmdLoypWq66703Oh3u0RLRPtFOEKQ0R4amThHde7SQu8cuD1Uu0UsrXfaTJmQAn5AkCzj1EW8dNWNBmyh/sdUDeBamSDo2VDgM50iHoYEzniSM7uHh3u0K7lhd2GUZ73llnhFBmbRMyRhNghM/ORD5pBBzCT8TJu3xONNxE0phqv6ZaTTfRChLKbGdgopPVUb/vfe9LZW+kjGjzayp3LEeQItq0hTSd6mrGXLmqczpu/mai294yVq8N4Mge9Z6WaTTHv4GCQ93PTlt56UtfmtpJbV4MFgyksn/pIVF2ANc2kC3xkiO0z7RU6i15IB3aaPHTPgmb3Uhyyn1aaXZb2mJuvzlcPktzpDhyXrPf7TnD1PRkJoTbE8dMC6PsaVPJeINQn49Dis0sUB4oV7K1sb4uFs3riFUTfEGS9c7V62NwSFxPMPS9vbPB6uruibkdHSPWhXuH2vF3th0yYRHChAAAEJFJREFU7Pj0dmgKGpOGbsRDG6OTJHw1zMk65G0slxv6SI19rPBT8ZxmZXuNMbNwrCVYtFQMU2l6CEaY0tjoxNgzaWy5Q0MG2FyJCwnRSLNgEo4/zxxU/8igclJ2tEo6XwKalk+8HOHqOaJCE5jfTxyuPdNxm1phtyKsMke2kDR+aDzEIww7FdoP+SIkCBMdh46AIEEgkfNagsXGI8dD8CBoSB0yCw/Tja4RK4TLSjDPhbPHEqJh1C+/E3FwnIo6PJE0J+NXedI2WGGI8Jx77rlJc4GcItDKlVbDofOEnakp9WwsZ7oKIUe4aUyUmTJEbqWlfBlNwyzXybHinC3PczuExXjccHIKfiPVz+H8a4863Wp7lbb6mussDVOt1mk8+eNH26wleuMNO1X+EHd1dFfnY6reZ7zxKD9E+dRTT02Dnwc84AHbDHq1sebGxljQ1hyrtmOh5vxmO2ZNjfMdQv1LfZ36MjH5OjU5GD6WWUWwNAAEy+gZwSLAjT4m6wid4YTLZOOdjuGNMKuYIRmwNNXGsNQoJi999kyjQz5dO9hfsWPKQprmhvNMeAaxtD5sMUwNaMTKjSqakOZP+sgWjQZyAn9k5bTTTksj4By3eIWXPhssJIsWCvFRB5AhU3PKTkecnfSkY6Rueoo2zFledRhsRGjKqgJVHuBg+hVxQ5xMfdKWslOAg8/G8EMzl504EM6JbMKZw+azd/HO8jCdHUwRZNhUFxbAEglSVtpkfg/aPNM3CBjHz2gOYdKZ251bOtJQtjQoNCQwVhesBEXWdIy7k1PXHFlzvDPeXb3UdsZyuczH8jcdn1tBSX7sboRdmWlfpl8NDM022JqDgwWC1drcFHvNaYqbe8j48ZceSdZQP/TR6PGHGtmn/Kj72r/zdHGzimCpEBo8w0rClRDWOSENk3XTqdAm+y6jhYeVjjI7mCINtAWubUlgRKMz08Bom+Ct4zzssMOSkTuyoxzYYzAmJ/Dhx8Cb9sioiEEq8mQagT2X+0iUNPhFfhAXU4n+m/fXyE0pmW5yTz6dTTHROtE+sX+y8zSSZKGD8DpyKn5xc/JlilAHzU5N2jpkm9PSvuSpkOy/ioXpjezYatFk7WinDnvP6e6UBy0dkq2slSuhrHNSBt6jiimbqWx7wxaoOm003LvSXiG4phpNEyLqnLpkWhlGOgQdIqf9724OvqYIixsZgap8G+66ek/HbTGLulu9P3Lss+sJOU72GjRZ6al9uUdTTw7XxUAcvagprlrdHX1pVl7fUTMQrPnruR3cN8C0advV5NuL3t0rVqb2P3/B9NpKY1YRLIVDwJiWInCNmAlxwn2yTsXaHRwtUlWDld+ZRgmhsppGw4IxEmOKkJ2b6TlkydSNTU51doiXqTfTi+wz8ooiq+zgqax8roPWSLpVZ6SEyJnu8Iydl31waLEIOkTQIgbLuMVv5CI+ZEta1MXSkA8kCvnznHOf0GCv514WnPLPuZf9phu7+MeU7XTKz0hwyGOeAlSHkFqEFMFCYt1TlmzcECS4O5SRaVxCezRn6o+WC4FDiE3xIvDioNXiqqSKjRaSLf7dxSFXu5vmbrxlq56oP/AhV1wj7eom3NQXC2PYVhow+m9wSBNuUIZY7K7O3lgWcFklDkdt3QKEP910U1z8iTOic1N9dA9E9KJcDc0Rza1R19oedc1tUd/SGnWm7Roao66xKeqbfHqrOZob1sTyDaujubE+fX6nsWlodTycfY7Hebxy74Ybboo1a9ZGx/wF4w6zM8pyxrOGTAYI0VwYiJVl8+b+NZapcOJWsWa702kNpwo3fWZqLJNVAorGxxQcUkNQ+X5a/qwGnGi37KVix2/2UwQb+7hMhvhhq2T1XW0nyJ6EcaWyhD1jV98yQzaUubKQT/6kk8s+n2uFYb6fy8//fC+f87Ppdp5JBEt5sq1TprlzyvVJmSFAluerSzSjVmcpY+VLCzma0wnSflqlpp7aA4mxu20ekGYDgCz8ncW7OzmEwYEwzBSX27E6kg/3tHH/XWc/3g3ZUQ8QbDKIZsmR7TvVgVzvEHt+HdmvuMgadQ9pQvbVPQMD9pquyadqugYKBoFMCaa7rNhR5W5gRHZraxZScdqg6f/TX/vKpE0eoJcaHDIH6ezuiVVr1saadetj9T0rY82addHT1Rlr16+LjRvWx6q77o7VK5bHV+5amsq0u7sryXbhUz1gd2s2Y/NWRQiZz+80tzRH+9y50T53TmrfymntuvVx+c9+EQsXIlfTiwTPSIKlMSFOdjm2kaUCUfkt+dbhY9o6fHu2jHcvLAWloRL4Ri8anzjzyJgWhJ/Z7hAW+NY6woehNlwIKBjna2Es6aWZQnYyWXImrNhv5a0aMmHK8dcSoer96jOCTVyO4qYvAsqcnZ72yTYHGVJPdPraF7KoczMdSBuKGFnxpjNUx0ZzOkP2WvZUo9FE2tUn7VK86ouOmdZMeuLdXTpEGJB52q52apob7qO5qjyDU/V/Dld7r/q/9hq5cSAziLQjk59MfAyyMuHJz52VH6d9kydZTkhDHXGfbNGpO/vPT5Y3yt7Br3ue8aM+Zjnkf657/OVnzt4/n2vrDHIlH57vrs67MxPJpiIZI9txeFaV1TCC1/57L0nnVE82E2U9qP+Dm8lzV2dnrFyxLO5ZdXdqu319vbHB3loDEd1dXbGp08rgDekbiOpSZ2dXdHUP1avu7p5YvXpNCHPg/vvFMcc8cEwzg51dfjOKYGmkpoWwaCNkjURBGw3TplgybEM86kzTDVaojUeDhaDZn8V+SuxGVBZpIVimwTR+Qmt3cKbevHutI7TyFJ9nCBaXGxqB5qh1nsOTYCtu+xDQac4k4W4fJ4b9NMim/gx8tFHfljNlrwPVyXondc0gyOKUseoI7RgbQFtyGEUzA9A5q2PacMZJWtJgI5fr5/YhP3NCeU/vTdNMVhkowgjG5CQHHzj5r5NDRh3ap7abCbC4hOM/+8nhhctYO+fDc/EI6xCfdJAaZJpcQVQQJGf2c+oGjTjbS/IlhxWP9PP/kc7ppXbCT05/JyQ1rZNQJrVuOJnPT8Js6KI2yDb/1Y0FCxfG4ODhqS5RgdmVfciOazDVwY0b1kVzi4F/b6xZvSp6erqjt6cnuro2RU93d/Sr0w2NsXjJ3tExb3yfldomEzvwz4wgWBo1QYww2adIwzV9RE1Ma0JrRX1sLx1TUgyYMWuN2pz6WA4JsxElMmUvJqvRjAaFJWQ0ftMarme7I+xgDXONZCQ32rORwpT724fAWGWxfbHuuFDIDe2Sz6aY0jP4YXhOs0FIa1uczldbRgzsg6ZjHc0Jy2bP6lMDLRpqZ9s/sA3RPsVhlG27DtM9u5ODI9MItkOICzIzkhuJNMBYfeNyG8/nfC+HzWdhHJkY5fNIaZf7BYEqAtV6NHR/60C9uSViTuXTN4v22Ly/GyI2OETw7XJqKlEdrNbVahq76rpuMLemXZWDcaRLrUwzZUpQ52+qgNGyvYqMoIxiOcKa8bV7hIul/6YKfG5hNGeXaQbURtVGbMIysqa5QtbEqwNAvBhxI1yz1TFmN0Xq/cfq8GYrBtPtvay0M51m6m0mOQMgu4Xbc4zwYxvjHmJk+oqmS9u2ASUtl3a2Pc7gSnhELk9lb088JUxBoCBQEJhKBLZPok1lDsaIC+HxUVFqb3YX9mBCdGibbOzIRoqAxRMRIiuNsFgkyLRfdVl9bVLC0F5ZSm7UzZDPqFqaNtVE0KhAES7aM53EDOCjta85of+mVk2TmgLcnQhWbbn6n0dWAMzPp9sIaUKFu5M9mwKiYTbg0Uat/tMu1TF1K2M5WY2HtmtzV3LilFNO2clvWZIrCBQECgLDIzDtCZZsm6dlZG7/G3tx+G/kS5tkWoCtBxKEdPmvM8wj5tE2wUOkqNSdHcgWNbszsuY7czoBGjKdhHhNSdCS5c5heFhn7l04WvXnY6+WuE+1y0Ql45n/O492nZ/ns3y5Vt7qgtVFDtcOZaYMXdNMOrP1oUVx36HMq/nIcTurP7VOmVfL3TV/DiTBgZAj6XBE7tUnR77HLgiRd2ZzJGyOp3rOacuLPOd6XU0/+5muZ3nNWim4wMG9qX4H8dFo02wPV27TFZ+Sr4JAQWB2IzDtCRaBSe3vy+3sqghpnY4OS6dDuJp2sAoJ8dJp6kx1Ygw9895GwxWjeJAyu0D70C/bDRto2sfJVKQOUxpVxxB0NjpYIB3OOiqfhmDkD0tYZAePTEyc83W+7wwjh2fZcNb9fLiHAIm3Goe0cwfputoZy1tOIz+TJx03DYazI5Mc99QRJIctjv86e0cmQ9VrYavpiVs6mYB55r/8Oudnntdi4F7GgeYG8WN3lHHx7o6MkXd25DjF531h76CJtW+X+5mw5PKYKedabKc63xnDqY63xFcQKAgUBLYXgWlPsAhm0wq+7m5TQdsy6GiQKUazuTOjqUK+sr2Ha45B7EhO3Faz0Db41Im4GXnbPPOKK65I6eSwOmBpjfUR1ex/pp117qZBae10/t7/rLPO2kJ4dGCZZLhGUpzhgsggMDCHpQPBdU/ZKSsLDhzZj2v+MmHIHXA+wy9fV8/D3c95428mOPXIkQlVPruXyZp7rm3IaQGGa/eKKwgUBAoCBYGZgcCMMHLX8SBX9lry3TEfiPXNORqtG2+8MWkIaCry/iugp42y7cJotkQ6LNNHdgKnYfCdPPHRPkgzO6QLSaBxYEeCOMw2532tPmKPBkvL5mmwkCAkKrtMdvL/ct6xCJgetLhDedhBv7iCQEGgIFAQmBkIzAiCBUpkiL2UlYM0Tddee23SPjFyN41ihE+TQUNi5aCPyF588cVJkzJWUZiqsgLpxz/+cVpBxwbr5z//eYpHugiGFU+IXf7O3lhxzsTn3tXBFSI1fUqwWibTJ1clJwWBgkBBoCAwGgIzhmB5CXYpiNDb3va29I07hq02LvQpAx8Gtr2AKSfaGNsunHzyyVumoEYDwbM8NSOsw9SM7+hJ88QTT0ybG5oKMzU2U6aixnrn8rwgUBAoCBQECgIFgR2DwIwiWCBAeBi101zZG4i2yfSdlYSmtKzOQpDs+oxsbQ8ZojGgFUPSaMve8Y53JLusHVMEJdaCQEGgIFAQKAgUBGYbAjOOYFULABFCppAo04NT7fLqubzqbKrjL/EVBAoCBYGCQEGgIDA7EZjRBGt2Fkl5q4JAQaAgUBAoCBQEZjoCU6/2memIlPwXBAoCBYGCQEGgIFAQmCQChWBNEsASvCBQECgIFAQKAgWBgkAtAoVg1SJS/hcECgIFgYJAQaAgUBCYJAKFYE0SwBK8IFAQKAgUBAoCBYGCQC0ChWDVIlL+FwQKAgWBgkBBoCBQEJgkAoVgTRLAErwgUBAoCBQECgIFgYJALQKFYNUiUv4XBAoCBYGCQEGgIFAQmCQChWBNEsASvCBQECgIFAQKAgWBgkAtAoVg1SJS/hcECgIFgYJAQaAgUBCYJAKFYE0SwBK8IFAQKAgUBAoCBYGCQC0ChWDVIlL+FwQKAgWBgkBBoCBQEJgkAoVgTRLAErwgUBAoCBQECgIFgYJALQKFYNUiUv4XBAoCBYGCQEGgIFAQmCQChWBNEsASvCBQECgIFAQKAgWBgkAtAoVg1SJS/hcECgIFgYJAQaAgUBCYJAL/H76QNs834Yb4AAAAAElFTkSuQmCC"
    }
   },
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "![image.png](attachment:image.png)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "NB: theese images are part of a coming post on that matter... "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "## Notebook settings\n",
    "%reload_ext autoreload\n",
    "%autoreload 2\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "## fastai imports\n",
    "from fastai.basics import *\n",
    "from fastai.data_block import ItemList\n",
    "from fastai.vision import *"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Import FastAI Audio modules here:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "from exp.nb_AudioCommon import *\n",
    "from exp.nb_DataBlock import *\n",
    "from exp.nb_DataAugmentation import *\n",
    "from exp.nb_AudioTransformsManager import *"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Other library imports"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "If this cell fails, you probably haven't installed the correct system/Python dependencies. You might need to re-run the \"System setup\" cell again."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch\n",
    "from pdb import set_trace\n",
    "from IPython.display import Audio\n",
    "from pydub import AudioSegment, silence\n",
    "import torchaudio\n",
    "from torchaudio import transforms\n",
    "import pandas as pd"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "#Options\n",
    "pd.set_option('display.max_colwidth', -1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Preparing the data"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We're going to do **phoneme classification** using the TIMIT dataset.\n",
    "\n",
    "The copyright on TIMIT isn't clear, and the data is hard to download, so we assume that the user has access to this dataset.\n",
    "\n",
    "**IMPORTANT:** the dataset should be located in the standard fast.ai location:\n",
    "> $HOME/.fastai/data/timit"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "#Fast.ai way to get folder for your dataset\n",
    "path = datapath4file('timit')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Parsing the data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "TEST PASED\n"
     ]
    }
   ],
   "source": [
    "def extractInformationsFromAbsolutePath(s: str):\n",
    "    items = list(filter(lambda x: len(x)>0,s.split('/')))\n",
    "    USAGE = items[0]\n",
    "    DIALECT = items[1]\n",
    "    SEX = items[2][0]\n",
    "    SPEAKER_ID = items[2][1:]\n",
    "    SENTENCE_ID = items[3]\n",
    "    return USAGE, DIALECT, SEX, SPEAKER_ID, SENTENCE_ID\n",
    "    \n",
    "def test_extractInformationsFromAbsolutePath():\n",
    "    USAGE, DIALECT, SEX, SPEAKER_ID, SENTENCE_ID = extractInformationsFromAbsolutePath('/TRAIN/DR7/MPAR0/SI2206')\n",
    "    assert 'TRAIN'==USAGE\n",
    "    assert 'DR7'==DIALECT\n",
    "    assert 'M'==SEX\n",
    "    assert 'PAR0'==SPEAKER_ID\n",
    "    assert 'SI2206'==SENTENCE_ID\n",
    "    print('TEST PASED')\n",
    "\n",
    "test_extractInformationsFromAbsolutePath()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>relPath</th>\n",
       "      <th>USAGE</th>\n",
       "      <th>DIALECT</th>\n",
       "      <th>SEX</th>\n",
       "      <th>SPEAKER_ID</th>\n",
       "      <th>SENTENCE_ID</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>IDX</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>MPAR0:SI2206</th>\n",
       "      <td>/TRAIN/DR7/MPAR0/SI2206</td>\n",
       "      <td>TRAIN</td>\n",
       "      <td>DR7</td>\n",
       "      <td>M</td>\n",
       "      <td>PAR0</td>\n",
       "      <td>SI2206</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>MPAR0:SX316</th>\n",
       "      <td>/TRAIN/DR7/MPAR0/SX316</td>\n",
       "      <td>TRAIN</td>\n",
       "      <td>DR7</td>\n",
       "      <td>M</td>\n",
       "      <td>PAR0</td>\n",
       "      <td>SX316</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>MPAR0:SI1576</th>\n",
       "      <td>/TRAIN/DR7/MPAR0/SI1576</td>\n",
       "      <td>TRAIN</td>\n",
       "      <td>DR7</td>\n",
       "      <td>M</td>\n",
       "      <td>PAR0</td>\n",
       "      <td>SI1576</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>MPAR0:SA1</th>\n",
       "      <td>/TRAIN/DR7/MPAR0/SA1</td>\n",
       "      <td>TRAIN</td>\n",
       "      <td>DR7</td>\n",
       "      <td>M</td>\n",
       "      <td>PAR0</td>\n",
       "      <td>SA1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>MPAR0:SX46</th>\n",
       "      <td>/TRAIN/DR7/MPAR0/SX46</td>\n",
       "      <td>TRAIN</td>\n",
       "      <td>DR7</td>\n",
       "      <td>M</td>\n",
       "      <td>PAR0</td>\n",
       "      <td>SX46</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                              relPath  USAGE DIALECT SEX SPEAKER_ID  \\\n",
       "IDX                                                                   \n",
       "MPAR0:SI2206  /TRAIN/DR7/MPAR0/SI2206  TRAIN  DR7     M   PAR0        \n",
       "MPAR0:SX316   /TRAIN/DR7/MPAR0/SX316   TRAIN  DR7     M   PAR0        \n",
       "MPAR0:SI1576  /TRAIN/DR7/MPAR0/SI1576  TRAIN  DR7     M   PAR0        \n",
       "MPAR0:SA1     /TRAIN/DR7/MPAR0/SA1     TRAIN  DR7     M   PAR0        \n",
       "MPAR0:SX46    /TRAIN/DR7/MPAR0/SX46    TRAIN  DR7     M   PAR0        \n",
       "\n",
       "             SENTENCE_ID  \n",
       "IDX                       \n",
       "MPAR0:SI2206  SI2206      \n",
       "MPAR0:SX316   SX316       \n",
       "MPAR0:SI1576  SI1576      \n",
       "MPAR0:SA1     SA1         \n",
       "MPAR0:SX46    SX46        "
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "def extractTimitDatasetInformation(path):\n",
    "    '''Put in a dataframe all the information about samples'''\n",
    "    import glob\n",
    "    files = glob.glob(str(path)+ '/**/*.WAV', recursive=True)\n",
    "    ret = pd.DataFrame(files, columns=['relPath'])\n",
    "    #Remove path and extension\n",
    "    ret['relPath'] = ret['relPath'].apply(lambda x: x.replace(str(path),'').replace('.WAV',''))\n",
    "\n",
    "    def extractIntoSeries(s):\n",
    "        USAGE, DIALECT, SEX, SPEAKER_ID, SENTENCE_ID = extractInformationsFromAbsolutePath(s)\n",
    "        return pd.Series({\n",
    "            'USAGE': USAGE,\n",
    "            'DIALECT': DIALECT,\n",
    "            'SEX': SEX,\n",
    "            'SPEAKER_ID': SPEAKER_ID,\n",
    "            'SENTENCE_ID': SENTENCE_ID,\n",
    "            'IDX': SEX + SPEAKER_ID + ':' + SENTENCE_ID})\n",
    "\n",
    "    #IMPORTANT: Apply multiple transformations at one\n",
    "    ret = ret.merge(ret['relPath'].apply(lambda s: extractIntoSeries(s)), left_index=True, right_index=True)    \n",
    "    ret = ret.set_index(['IDX']) # Add index\n",
    "    return ret\n",
    "                \n",
    "all_files_df = extractTimitDatasetInformation(path)\n",
    "all_files_df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Number of files: 6300\n"
     ]
    }
   ],
   "source": [
    "print(f'Number of files: {len(all_files_df)}')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### OPTIONAL Decimate dataset\n",
    "To speedup computation we reduce train/valid set."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "USAGE\n",
       "TEST     168\n",
       "TRAIN    462\n",
       "Name: relPath, dtype: int64"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#DON'T RUN THIS CELL IN YOU WANT TOI USE ALL DATA\n",
    "def decimate(train_df:pd.DataFrame, valid_df:pd.DataFrame, pct=.1)->pd.DataFrame:\n",
    "    t1=train_df.sample(frac=pct, random_state=1)\n",
    "    t2=valid_df.sample(frac=pct, random_state=1)\n",
    "    return pd.concat([t1,t2],axis=0)\n",
    "\n",
    "all_files_df=decimate(all_files_df[all_files_df.USAGE=='TRAIN'],all_files_df[all_files_df.USAGE=='TEST'],.1)\n",
    "all_files_df.groupby(['USAGE'])['relPath'].count()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Phoneme classification"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Create data bunch"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "def process_phn_file(idx:str, phn_file, sig, sr, ext, delimiter=' ')->pd.DataFrame:\n",
    "    df = pd.read_csv(phn_file, delimiter=delimiter, header=None)\n",
    "    df.columns = ['START', 'END', ext]\n",
    "    df['SIG'] = df.apply(lambda x : torch.Tensor(sig[x['START']: x['END']]), axis=1)\n",
    "    df['SR'] = sr\n",
    "    df['IDX'] = idx\n",
    "    df['IDXN'] = pd.Series(map(lambda x: idx+':'+str(x), range(len(df))))   \n",
    "    df = df.set_index(['IDXN'])\n",
    "    return df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "/home/ste/.fastai/data/timit/TRAIN/DR2/MJMD0/SX308.WAV \tExists: True\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>START</th>\n",
       "      <th>END</th>\n",
       "      <th>PHN</th>\n",
       "      <th>SIG</th>\n",
       "      <th>SR</th>\n",
       "      <th>IDX</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>IDXN</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>MJMD0:SX308:0</th>\n",
       "      <td>0</td>\n",
       "      <td>2320</td>\n",
       "      <td>h#</td>\n",
       "      <td>[tensor(0.0001), tensor(0.0002), tensor(-3.0518e-05), tensor(6.1035e-05), tensor(-3.0518e-05), tensor(6.1035e-05), tensor(9.1553e-05), tensor(0.0001), tensor(0.0001), tensor(3.0518e-05), tensor(9.1553e-05), tensor(0.), tensor(-3.0518e-05), tensor(-9.1553e-05), tensor(0.0002), tensor(0.0001), tensor(0.0002), tensor(9.1553e-05), tensor(6.1035e-05), tensor(9.1553e-05), tensor(0.), tensor(0.0002), tensor(9.1553e-05), tensor(0.0001), tensor(0.0002), tensor(0.0002), tensor(0.0002), tensor(0.0002), tensor(0.0001), tensor(6.1035e-05), tensor(9.1553e-05), tensor(-9.1553e-05), tensor(3.0518e-05), tensor(0.0002), tensor(0.0002), tensor(0.0002), tensor(6.1035e-05), tensor(0.0002), tensor(0.), tensor(0.0001), tensor(9.1553e-05), tensor(0.0002), tensor(0.0002), tensor(0.0002), tensor(0.0001), tensor(0.0002), tensor(6.1035e-05), tensor(9.1553e-05), tensor(6.1035e-05), tensor(6.1035e-05), tensor(0.0002), tensor(0.), tensor(0.0002), tensor(0.0002), tensor(0.0002), tensor(9.1553e-05), tensor(9.1553e-05), tensor(9.1553e-05), tensor(0.0001), tensor(0.0001), tensor(0.0002), tensor(0.0002), tensor(0.0001), tensor(9.1553e-05), tensor(3.0518e-05), tensor(9.1553e-05), tensor(-6.1035e-05), tensor(-6.1035e-05), tensor(6.1035e-05), tensor(0.0001), tensor(3.0518e-05), tensor(3.0518e-05), tensor(6.1035e-05), tensor(3.0518e-05), tensor(-6.1035e-05), tensor(-3.0518e-05), tensor(3.0518e-05), tensor(0.0001), tensor(-3.0518e-05), tensor(0.0001), tensor(3.0518e-05), tensor(0.0001), tensor(0.), tensor(3.0518e-05), tensor(0.), tensor(0.), tensor(9.1553e-05), tensor(9.1553e-05), tensor(0.0001), tensor(0.0002), tensor(9.1553e-05), tensor(3.0518e-05), tensor(0.0002), tensor(0.), tensor(9.1553e-05), tensor(9.1553e-05), tensor(0.0001), tensor(0.0002), tensor(0.0001), tensor(9.1553e-05), ...]</td>\n",
       "      <td>16000</td>\n",
       "      <td>MJMD0:SX308</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>MJMD0:SX308:1</th>\n",
       "      <td>2320</td>\n",
       "      <td>4803</td>\n",
       "      <td>sh</td>\n",
       "      <td>[tensor(3.0518e-05), tensor(-0.0002), tensor(-3.0518e-05), tensor(0.0004), tensor(0.0003), tensor(0.0002), tensor(-0.0002), tensor(6.1035e-05), tensor(0.0002), tensor(0.0002), tensor(9.1553e-05), tensor(-0.0004), tensor(0.0002), tensor(0.0006), tensor(0.), tensor(-0.0002), tensor(-6.1035e-05), tensor(9.1553e-05), tensor(0.0003), tensor(0.0003), tensor(0.0002), tensor(-0.0005), tensor(-0.0005), tensor(0.0001), tensor(0.0003), tensor(0.0005), tensor(0.0005), tensor(-0.0007), tensor(-0.0005), tensor(0.0007), tensor(0.0005), tensor(-0.0002), tensor(-0.0002), tensor(0.0003), tensor(-0.0001), tensor(-0.0005), tensor(0.0005), tensor(0.0006), tensor(-0.0004), tensor(-0.0003), tensor(-3.0518e-05), tensor(-0.0002), tensor(0.0007), tensor(0.0005), tensor(-0.0007), tensor(-0.0006), tensor(-0.0002), tensor(0.0005), tensor(0.0008), tensor(-0.0002), tensor(-0.0008), tensor(-9.1553e-05), tensor(0.0006), tensor(0.0006), tensor(0.0002), tensor(-0.0003), tensor(-3.0518e-05), tensor(6.1035e-05), tensor(-0.0005), tensor(-0.0003), tensor(3.0518e-05), tensor(-0.0004), tensor(-0.0005), tensor(-6.1035e-05), tensor(9.1553e-05), tensor(0.0003), tensor(0.0006), tensor(-0.0005), tensor(-0.0005), tensor(0.), tensor(-0.0002), tensor(0.0004), tensor(0.0005), tensor(-0.0003), tensor(-0.0003), tensor(0.0003), tensor(0.0006), tensor(0.0005), tensor(-0.0002), tensor(-0.0005), tensor(-3.0518e-05), tensor(-3.0518e-05), tensor(0.0005), tensor(-0.0004), tensor(-0.0005), tensor(0.0002), tensor(0.0002), tensor(0.0006), tensor(0.0003), tensor(-0.0004), tensor(-0.0003), tensor(0.0003), tensor(0.0004), tensor(0.0001), tensor(-0.0011), tensor(-0.0003), tensor(0.0006), tensor(0.0005), tensor(0.0009), tensor(0.0004), ...]</td>\n",
       "      <td>16000</td>\n",
       "      <td>MJMD0:SX308</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "               START   END PHN  \\\n",
       "IDXN                             \n",
       "MJMD0:SX308:0  0      2320  h#   \n",
       "MJMD0:SX308:1  2320   4803  sh   \n",
       "\n",
       "                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                               SIG  \\\n",
       "IDXN                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                 \n",
       "MJMD0:SX308:0  [tensor(0.0001), tensor(0.0002), tensor(-3.0518e-05), tensor(6.1035e-05), tensor(-3.0518e-05), tensor(6.1035e-05), tensor(9.1553e-05), tensor(0.0001), tensor(0.0001), tensor(3.0518e-05), tensor(9.1553e-05), tensor(0.), tensor(-3.0518e-05), tensor(-9.1553e-05), tensor(0.0002), tensor(0.0001), tensor(0.0002), tensor(9.1553e-05), tensor(6.1035e-05), tensor(9.1553e-05), tensor(0.), tensor(0.0002), tensor(9.1553e-05), tensor(0.0001), tensor(0.0002), tensor(0.0002), tensor(0.0002), tensor(0.0002), tensor(0.0001), tensor(6.1035e-05), tensor(9.1553e-05), tensor(-9.1553e-05), tensor(3.0518e-05), tensor(0.0002), tensor(0.0002), tensor(0.0002), tensor(6.1035e-05), tensor(0.0002), tensor(0.), tensor(0.0001), tensor(9.1553e-05), tensor(0.0002), tensor(0.0002), tensor(0.0002), tensor(0.0001), tensor(0.0002), tensor(6.1035e-05), tensor(9.1553e-05), tensor(6.1035e-05), tensor(6.1035e-05), tensor(0.0002), tensor(0.), tensor(0.0002), tensor(0.0002), tensor(0.0002), tensor(9.1553e-05), tensor(9.1553e-05), tensor(9.1553e-05), tensor(0.0001), tensor(0.0001), tensor(0.0002), tensor(0.0002), tensor(0.0001), tensor(9.1553e-05), tensor(3.0518e-05), tensor(9.1553e-05), tensor(-6.1035e-05), tensor(-6.1035e-05), tensor(6.1035e-05), tensor(0.0001), tensor(3.0518e-05), tensor(3.0518e-05), tensor(6.1035e-05), tensor(3.0518e-05), tensor(-6.1035e-05), tensor(-3.0518e-05), tensor(3.0518e-05), tensor(0.0001), tensor(-3.0518e-05), tensor(0.0001), tensor(3.0518e-05), tensor(0.0001), tensor(0.), tensor(3.0518e-05), tensor(0.), tensor(0.), tensor(9.1553e-05), tensor(9.1553e-05), tensor(0.0001), tensor(0.0002), tensor(9.1553e-05), tensor(3.0518e-05), tensor(0.0002), tensor(0.), tensor(9.1553e-05), tensor(9.1553e-05), tensor(0.0001), tensor(0.0002), tensor(0.0001), tensor(9.1553e-05), ...]   \n",
       "MJMD0:SX308:1  [tensor(3.0518e-05), tensor(-0.0002), tensor(-3.0518e-05), tensor(0.0004), tensor(0.0003), tensor(0.0002), tensor(-0.0002), tensor(6.1035e-05), tensor(0.0002), tensor(0.0002), tensor(9.1553e-05), tensor(-0.0004), tensor(0.0002), tensor(0.0006), tensor(0.), tensor(-0.0002), tensor(-6.1035e-05), tensor(9.1553e-05), tensor(0.0003), tensor(0.0003), tensor(0.0002), tensor(-0.0005), tensor(-0.0005), tensor(0.0001), tensor(0.0003), tensor(0.0005), tensor(0.0005), tensor(-0.0007), tensor(-0.0005), tensor(0.0007), tensor(0.0005), tensor(-0.0002), tensor(-0.0002), tensor(0.0003), tensor(-0.0001), tensor(-0.0005), tensor(0.0005), tensor(0.0006), tensor(-0.0004), tensor(-0.0003), tensor(-3.0518e-05), tensor(-0.0002), tensor(0.0007), tensor(0.0005), tensor(-0.0007), tensor(-0.0006), tensor(-0.0002), tensor(0.0005), tensor(0.0008), tensor(-0.0002), tensor(-0.0008), tensor(-9.1553e-05), tensor(0.0006), tensor(0.0006), tensor(0.0002), tensor(-0.0003), tensor(-3.0518e-05), tensor(6.1035e-05), tensor(-0.0005), tensor(-0.0003), tensor(3.0518e-05), tensor(-0.0004), tensor(-0.0005), tensor(-6.1035e-05), tensor(9.1553e-05), tensor(0.0003), tensor(0.0006), tensor(-0.0005), tensor(-0.0005), tensor(0.), tensor(-0.0002), tensor(0.0004), tensor(0.0005), tensor(-0.0003), tensor(-0.0003), tensor(0.0003), tensor(0.0006), tensor(0.0005), tensor(-0.0002), tensor(-0.0005), tensor(-3.0518e-05), tensor(-3.0518e-05), tensor(0.0005), tensor(-0.0004), tensor(-0.0005), tensor(0.0002), tensor(0.0002), tensor(0.0006), tensor(0.0003), tensor(-0.0004), tensor(-0.0003), tensor(0.0003), tensor(0.0004), tensor(0.0001), tensor(-0.0011), tensor(-0.0003), tensor(0.0006), tensor(0.0005), tensor(0.0009), tensor(0.0004), ...]                                                                               \n",
       "\n",
       "                  SR          IDX  \n",
       "IDXN                               \n",
       "MJMD0:SX308:0  16000  MJMD0:SX308  \n",
       "MJMD0:SX308:1  16000  MJMD0:SX308  "
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Number of rows: 34\n",
      "TEST DONE\n"
     ]
    }
   ],
   "source": [
    "def extractItems(path, idx, relativePath, ext, verbose=False)->pd.DataFrame:\n",
    "    base_path = str(path) + relativePath\n",
    "    wav_file = base_path + \".WAV\"\n",
    "    wav_exists = os.path.isfile(wav_file)\n",
    "    if verbose: print(wav_file, f'\\tExists: {wav_exists}')\n",
    "    #Sanity check\n",
    "    if not wav_exists: return none\n",
    "    #Proceed\n",
    "    sigTwoChannels,sr = torchaudio.load(wav_file)\n",
    "    sig = sigTwoChannels[0] # MONO NO STEREO - convert to \"one dimension\"\n",
    "    phn_file = base_path + '.' + ext\n",
    "            \n",
    "    ret = process_phn_file(idx, phn_file, sig, sr, ext, delimiter=' ')\n",
    "    \n",
    "    return ret\n",
    "            \n",
    "def test_extractItems_PHN():\n",
    "    row = all_files_df.iloc[0]\n",
    "    ret_df = extractItems(path, row.name, row['relPath'], 'PHN', verbose=True)\n",
    "    display(ret_df.head(2))\n",
    "    print(f'Number of rows: {len(ret_df)}')\n",
    "#    assert 54==len(ret_df)\n",
    "    print('TEST DONE')\n",
    "\n",
    "test_extractItems_PHN()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "/home/ste/.fastai/data/timit/TRAIN/DR2/MJMD0/SX308.WAV \tExists: True\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>START</th>\n",
       "      <th>END</th>\n",
       "      <th>WRD</th>\n",
       "      <th>SIG</th>\n",
       "      <th>SR</th>\n",
       "      <th>IDX</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>IDXN</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>MJMD0:SX308:0</th>\n",
       "      <td>2320</td>\n",
       "      <td>11400</td>\n",
       "      <td>shipbuilding</td>\n",
       "      <td>[tensor(3.0518e-05), tensor(-0.0002), tensor(-3.0518e-05), tensor(0.0004), tensor(0.0003), tensor(0.0002), tensor(-0.0002), tensor(6.1035e-05), tensor(0.0002), tensor(0.0002), tensor(9.1553e-05), tensor(-0.0004), tensor(0.0002), tensor(0.0006), tensor(0.), tensor(-0.0002), tensor(-6.1035e-05), tensor(9.1553e-05), tensor(0.0003), tensor(0.0003), tensor(0.0002), tensor(-0.0005), tensor(-0.0005), tensor(0.0001), tensor(0.0003), tensor(0.0005), tensor(0.0005), tensor(-0.0007), tensor(-0.0005), tensor(0.0007), tensor(0.0005), tensor(-0.0002), tensor(-0.0002), tensor(0.0003), tensor(-0.0001), tensor(-0.0005), tensor(0.0005), tensor(0.0006), tensor(-0.0004), tensor(-0.0003), tensor(-3.0518e-05), tensor(-0.0002), tensor(0.0007), tensor(0.0005), tensor(-0.0007), tensor(-0.0006), tensor(-0.0002), tensor(0.0005), tensor(0.0008), tensor(-0.0002), tensor(-0.0008), tensor(-9.1553e-05), tensor(0.0006), tensor(0.0006), tensor(0.0002), tensor(-0.0003), tensor(-3.0518e-05), tensor(6.1035e-05), tensor(-0.0005), tensor(-0.0003), tensor(3.0518e-05), tensor(-0.0004), tensor(-0.0005), tensor(-6.1035e-05), tensor(9.1553e-05), tensor(0.0003), tensor(0.0006), tensor(-0.0005), tensor(-0.0005), tensor(0.), tensor(-0.0002), tensor(0.0004), tensor(0.0005), tensor(-0.0003), tensor(-0.0003), tensor(0.0003), tensor(0.0006), tensor(0.0005), tensor(-0.0002), tensor(-0.0005), tensor(-3.0518e-05), tensor(-3.0518e-05), tensor(0.0005), tensor(-0.0004), tensor(-0.0005), tensor(0.0002), tensor(0.0002), tensor(0.0006), tensor(0.0003), tensor(-0.0004), tensor(-0.0003), tensor(0.0003), tensor(0.0004), tensor(0.0001), tensor(-0.0011), tensor(-0.0003), tensor(0.0006), tensor(0.0005), tensor(0.0009), tensor(0.0004), ...]</td>\n",
       "      <td>16000</td>\n",
       "      <td>MJMD0:SX308</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "               START    END           WRD  \\\n",
       "IDXN                                        \n",
       "MJMD0:SX308:0  2320   11400  shipbuilding   \n",
       "\n",
       "                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                   SIG  \\\n",
       "IDXN                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                     \n",
       "MJMD0:SX308:0  [tensor(3.0518e-05), tensor(-0.0002), tensor(-3.0518e-05), tensor(0.0004), tensor(0.0003), tensor(0.0002), tensor(-0.0002), tensor(6.1035e-05), tensor(0.0002), tensor(0.0002), tensor(9.1553e-05), tensor(-0.0004), tensor(0.0002), tensor(0.0006), tensor(0.), tensor(-0.0002), tensor(-6.1035e-05), tensor(9.1553e-05), tensor(0.0003), tensor(0.0003), tensor(0.0002), tensor(-0.0005), tensor(-0.0005), tensor(0.0001), tensor(0.0003), tensor(0.0005), tensor(0.0005), tensor(-0.0007), tensor(-0.0005), tensor(0.0007), tensor(0.0005), tensor(-0.0002), tensor(-0.0002), tensor(0.0003), tensor(-0.0001), tensor(-0.0005), tensor(0.0005), tensor(0.0006), tensor(-0.0004), tensor(-0.0003), tensor(-3.0518e-05), tensor(-0.0002), tensor(0.0007), tensor(0.0005), tensor(-0.0007), tensor(-0.0006), tensor(-0.0002), tensor(0.0005), tensor(0.0008), tensor(-0.0002), tensor(-0.0008), tensor(-9.1553e-05), tensor(0.0006), tensor(0.0006), tensor(0.0002), tensor(-0.0003), tensor(-3.0518e-05), tensor(6.1035e-05), tensor(-0.0005), tensor(-0.0003), tensor(3.0518e-05), tensor(-0.0004), tensor(-0.0005), tensor(-6.1035e-05), tensor(9.1553e-05), tensor(0.0003), tensor(0.0006), tensor(-0.0005), tensor(-0.0005), tensor(0.), tensor(-0.0002), tensor(0.0004), tensor(0.0005), tensor(-0.0003), tensor(-0.0003), tensor(0.0003), tensor(0.0006), tensor(0.0005), tensor(-0.0002), tensor(-0.0005), tensor(-3.0518e-05), tensor(-3.0518e-05), tensor(0.0005), tensor(-0.0004), tensor(-0.0005), tensor(0.0002), tensor(0.0002), tensor(0.0006), tensor(0.0003), tensor(-0.0004), tensor(-0.0003), tensor(0.0003), tensor(0.0004), tensor(0.0001), tensor(-0.0011), tensor(-0.0003), tensor(0.0006), tensor(0.0005), tensor(0.0009), tensor(0.0004), ...]   \n",
       "\n",
       "                  SR          IDX  \n",
       "IDXN                               \n",
       "MJMD0:SX308:0  16000  MJMD0:SX308  "
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Number of rows: 6\n",
      "TEST DONE\n"
     ]
    }
   ],
   "source": [
    "def test_extractItems_WRD():\n",
    "    row = all_files_df.iloc[0]\n",
    "    ret_df = extractItems(path, row.name, row['relPath'], 'WRD', verbose=True)\n",
    "    display(ret_df.head(1))\n",
    "    print(f'Number of rows: {len(ret_df)}')\n",
    "#    assert 14==len(ret_df)\n",
    "    print('TEST DONE')\n",
    "\n",
    "test_extractItems_WRD()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>relPath</th>\n",
       "      <th>USAGE</th>\n",
       "      <th>DIALECT</th>\n",
       "      <th>SEX</th>\n",
       "      <th>SPEAKER_ID</th>\n",
       "      <th>SENTENCE_ID</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>IDX</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>MJMD0:SX308</th>\n",
       "      <td>/TRAIN/DR2/MJMD0/SX308</td>\n",
       "      <td>TRAIN</td>\n",
       "      <td>DR2</td>\n",
       "      <td>M</td>\n",
       "      <td>JMD0</td>\n",
       "      <td>SX308</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>FLEH0:SI2311</th>\n",
       "      <td>/TRAIN/DR7/FLEH0/SI2311</td>\n",
       "      <td>TRAIN</td>\n",
       "      <td>DR7</td>\n",
       "      <td>F</td>\n",
       "      <td>LEH0</td>\n",
       "      <td>SI2311</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                              relPath  USAGE DIALECT SEX SPEAKER_ID  \\\n",
       "IDX                                                                   \n",
       "MJMD0:SX308   /TRAIN/DR2/MJMD0/SX308   TRAIN  DR2     M   JMD0        \n",
       "FLEH0:SI2311  /TRAIN/DR7/FLEH0/SI2311  TRAIN  DR7     F   LEH0        \n",
       "\n",
       "             SENTENCE_ID  \n",
       "IDX                       \n",
       "MJMD0:SX308   SX308       \n",
       "FLEH0:SI2311  SI2311      "
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "all_files_df.head(2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<class 'pandas.core.frame.DataFrame'>\n",
      "Index: 630 entries, MJMD0:SX308 to FCAU0:SX317\n",
      "Data columns (total 6 columns):\n",
      "relPath        630 non-null object\n",
      "USAGE          630 non-null object\n",
      "DIALECT        630 non-null object\n",
      "SEX            630 non-null object\n",
      "SPEAKER_ID     630 non-null object\n",
      "SENTENCE_ID    630 non-null object\n",
      "dtypes: object(6)\n",
      "memory usage: 34.5+ KB\n"
     ]
    }
   ],
   "source": [
    "all_files_df.info()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Rows: 376\n",
      "<class 'pandas.core.frame.DataFrame'>\n",
      "Index: 376 entries, MJMD0:SX308:0 to MADD0:SI538:36\n",
      "Data columns (total 6 columns):\n",
      "START    376 non-null int64\n",
      "END      376 non-null int64\n",
      "PHN      376 non-null object\n",
      "SIG      376 non-null object\n",
      "SR       376 non-null int64\n",
      "IDX      376 non-null object\n",
      "dtypes: int64(3), object(3)\n",
      "memory usage: 20.6+ KB\n"
     ]
    }
   ],
   "source": [
    "def createItemsDataFrame(path, files_df, ext):\n",
    "    ret = pd.DataFrame()\n",
    "    for i,row in files_df.iterrows():\n",
    "        tmp_df = extractItems(path, row.name, row['relPath'], ext)\n",
    "        ret = ret.append(tmp_df)\n",
    "        #print(len(ret))\n",
    "#    ret = files_df.apply(lambda row: ret.append(extractWords(path, row.name, row['relPath']), ignore_index=True), axis=1)\n",
    "#    files_df.apply(lambda row: print(row), axis=1)\n",
    "    return ret\n",
    "\n",
    "def test_createItemsDataFrame():\n",
    "    tdf = createItemsDataFrame(path, all_files_df.head(10), 'PHN')\n",
    "    print(f'Rows: {len(tdf)}')\n",
    "    tdf.info()\n",
    "#    assert 86==len(tdf)\n",
    "\n",
    "test_createItemsDataFrame()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 28.3 s, sys: 523 ms, total: 28.8 s\n",
      "Wall time: 4.81 s\n",
      "<class 'pandas.core.frame.DataFrame'>\n",
      "Index: 24034 entries, MJMD0:SX308:0 to FCAU0:SX317:38\n",
      "Data columns (total 6 columns):\n",
      "START    24034 non-null int64\n",
      "END      24034 non-null int64\n",
      "PHN      24034 non-null object\n",
      "SIG      24034 non-null object\n",
      "SR       24034 non-null int64\n",
      "IDX      24034 non-null object\n",
      "dtypes: int64(3), object(3)\n",
      "memory usage: 1.3+ MB\n"
     ]
    }
   ],
   "source": [
    "%time all_phonemes_df = createItemsDataFrame(path, all_files_df, 'PHN')\n",
    "all_phonemes_df.info()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Phonemes count: 24034\n"
     ]
    }
   ],
   "source": [
    "print(f'Phonemes count: {len(all_phonemes_df)}')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Join data and create \"src\" "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "#Join files with words\n",
    "src = all_phonemes_df.merge(all_files_df,left_on='IDX',right_on='IDX')\n",
    "assert len(src)==len(all_phonemes_df)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "USAGE\n",
       "TEST     6367 \n",
       "TRAIN    17667\n",
       "Name: SIG, dtype: int64"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#Optionally subsample and ensure good distribution among train and \"valid\"\n",
    "#src = src.sample(5000, random_state=1) # Uncomment this to reduce dataset\n",
    "src.groupby(['USAGE'])['SIG'].count()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>PHN</th>\n",
       "      <th>FREQ</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>aa</td>\n",
       "      <td>402</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>ae</td>\n",
       "      <td>548</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>ah</td>\n",
       "      <td>316</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>ao</td>\n",
       "      <td>419</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>aw</td>\n",
       "      <td>75</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "  PHN  FREQ\n",
       "0  aa  402 \n",
       "1  ae  548 \n",
       "2  ah  316 \n",
       "3  ao  419 \n",
       "4  aw  75  "
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "wrd_freq = src.groupby(['PHN'])['SIG'].count().reset_index(name='FREQ')\n",
    "wrd_freq.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "#Add word frequency measure to src\n",
    "src = src.merge(wrd_freq,left_on='PHN',right_on='PHN')\n",
    "assert len(src)==len(all_phonemes_df) # does not pass if src is reduced"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "24034\n"
     ]
    }
   ],
   "source": [
    "# Keep high frequency words\n",
    "#src = src[src.FREQ>=8]\n",
    "print(len(src))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(tensor([ 1.2207e-04,  1.5259e-04, -3.0518e-05,  ..., -3.6621e-04,\n",
       "          4.8828e-04,  6.1035e-04]), 16000)"
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#Create merged column...\n",
    "src['SampleAndSr'] = src.apply(lambda row: (row.SIG,row.SR),axis=1)\n",
    "src.iloc[0]['SampleAndSr']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "is_valid\n",
       "False    17667\n",
       "True     6367 \n",
       "Name: SIG, dtype: int64"
      ]
     },
     "execution_count": 25,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#Create validation set column from \"test\" hold out\n",
    "src['is_valid'] = src['USAGE'].apply(lambda x: not (x=='TRAIN'))\n",
    "src.groupby(['is_valid'])['SIG'].count()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "61"
      ]
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#Count unique words\n",
    "len(src.PHN.unique())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Useful statistics"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Shortest item: 95\n",
      "Longest item: 74285\n"
     ]
    }
   ],
   "source": [
    "src['LENGHT'] = src.SIG.apply(lambda x: x.shape[0])\n",
    "print(f'Shortest item: {src.LENGHT.min()}')\n",
    "print(f'Longest item: {src.LENGHT.max()}')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.axes._subplots.AxesSubplot at 0x7f9587a1c7f0>"
      ]
     },
     "execution_count": 28,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "src['LENGHT'].hist(bins=61)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**WARNING** there are zero lenght items that should be filtered..."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Shortest item: 95\n",
      "Longest item: 9971\n",
      "Number of items: 24000\n"
     ]
    }
   ],
   "source": [
    "#Discarding too short and too long sounds\n",
    "min_duration = 50 # in samples\n",
    "max_duration = 10000\n",
    "src = src[src.LENGHT>min_duration]\n",
    "src = src[src.LENGHT<=max_duration]\n",
    "print(f'Shortest item: {src.LENGHT.min()}')\n",
    "print(f'Longest item: {src.LENGHT.max()}')\n",
    "print(f'Number of items: {len(src)}')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Classify phonemes"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [],
   "source": [
    "samples = src.sample(10,random_state=1)\n",
    "adsAndLbl = [(AudioData(samples.iloc[i].SIG,samples.iloc[i].SR),samples.iloc[i].PHN) for i in range(len(samples))]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "10"
      ]
     },
     "execution_count": 31,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(adsAndLbl)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "\n",
       "                <audio controls=\"controls\" >\n",
       "                    <source src=\"data:audio/wav;base64,UklGRlwBAABXQVZFZm10IBAAAAABAAEAgD4AAAB9AAACABAAZGF0YTgBAACSAXICZgElA3ICsQe7Fv9/jXZZx8DhF0luFyUjaFMY6Znl+1STepEIDKi7/V3yT/jf5z2zQd6ER6E0D9q30sypDsEC+ZnFV5Urrt6u7vrjK5nlWAeK8mYBgw42LxQXIviEBz8+JUOqKnoY9/AAAFlAJjzwLLv9wrqP9rMAqdhdshW+Eqz6wuLZOagApy7gAtnK8Bb3atN55tEGYw9dK+UkFwmRCD4lVDV9KiwgZxrRJgMrrRxRCiUD9fd4BvYQQf4s507fmeUk8VDxuOtk7yXqse7J997uF/CO/TkBgw6DDnX7G/sSBd4HgRVIFJ4JkgG1+d8ARQKI+Qj9L/ns6PPlp98i2Pvbddui267citLyzFjOeM3D0+nWJNGzx5W69rfJt5C2Gcl90W3F19EP2si+ytCu/M778QU=\" type=\"audio/wav\" />\n",
       "                    Your browser does not support the audio element.\n",
       "                </audio>\n",
       "              "
      ],
      "text/plain": [
       "<IPython.lib.display.Audio object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "s = adsAndLbl[4][0]\n",
    "display(Audio(s.sig, rate=s.sr))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "z tensor([-0.0024,  0.0034,  0.0006, -0.0034, -0.0013])\n",
      "eh tensor([0.0106, 0.0108, 0.0137, 0.0164, 0.0186])\n",
      "s tensor([-0.0008, -0.0005, -0.0003, -0.0002,  0.0000])\n",
      "ix tensor([-0.0019, -0.0039,  0.0016,  0.0028, -0.0034])\n",
      "d tensor([0.0003, 0.0004, 0.0002, 0.0005, 0.0004])\n",
      "h# tensor([ 0.0000e+00,  9.1553e-05,  1.2207e-04,  3.0518e-05, -6.1035e-05])\n",
      "em tensor([-0.0006,  0.0014,  0.0038,  0.0054,  0.0060])\n",
      "v tensor([-0.0017, -0.0013, -0.0023, -0.0019, -0.0019])\n",
      "h# tensor([3.0518e-05, 6.1035e-05, 3.0518e-05, 0.0000e+00, 9.1553e-05])\n",
      "el tensor([0.0081, 0.0100, 0.0122, 0.0150, 0.0180])\n"
     ]
    }
   ],
   "source": [
    "for ad in adsAndLbl[:10]:\n",
    "    print(ad[1],ad[0].sig[:5])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Create transformations"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [],
   "source": [
    "#Data Augmentetion\n",
    "tm = AudioTfmsManager.get_audio_tfms_manager(\n",
    "                            spectro=True, #We're going to replace it...\n",
    "                            mx_to_pad=max_duration, #1 sec window\n",
    "                            white_noise=True, noise_scl=1e-4, # Small noise\n",
    "                            modulate_volume=True, lower_gain=.90, upper_gain=1.05, # Not big volume variation\n",
    "                            random_cutout=False,\n",
    "                            pad_with_silence=False,\n",
    "                            pitch_warp=False,\n",
    "                            down_and_up=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "8ff30aef0c204c41b1562d4dacff2309",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Tab(children=(Output(), VBox(children=(Label(value='(0)\\tfunctools.partial(<function tfm_add_white_noise at 0x…"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "<exp.nb_AudioCommon.AudioData at 0x7f9587a046a0>"
      ]
     },
     "execution_count": 35,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sl=adsAndLbl[1]\n",
    "tm.try_data_augmentation(ad=sl[0],lbl=sl[1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[[functools.partial(<function tfm_add_white_noise at 0x7f95880679d8>, noise_scl=0.0001, lower_gain=0.9, upper_gain=1.05),\n",
       "  functools.partial(<function tfm_modulate_volume at 0x7f9588067a60>, noise_scl=0.0001, lower_gain=0.9, upper_gain=1.05),\n",
       "  functools.partial(<function tfm_pad_to_max at 0x7f9588067d90>, mx=10000),\n",
       "  functools.partial(<function tfm_spectro at 0x7f95883c3158>, noise_scl=0.0001, lower_gain=0.9, upper_gain=1.05)],\n",
       " [functools.partial(<function tfm_pad_to_max at 0x7f9588067d90>, mx=10000),\n",
       "  functools.partial(<function tfm_spectro at 0x7f95883c3158>, noise_scl=0.0001, lower_gain=0.9, upper_gain=1.05)]]"
      ]
     },
     "execution_count": 36,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tm.get_tfms()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "([[functools.partial(<function tfm_spectro at 0x7f95883c3158>, noise_scl=0.0001, lower_gain=0.9, upper_gain=1.05)]],\n",
       " [[functools.partial(<function tfm_spectro at 0x7f95883c3158>, noise_scl=0.0001, lower_gain=0.9, upper_gain=1.05)]])"
      ]
     },
     "execution_count": 37,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#This is the \"POST\" transformation group that we're going to replace, both on train and validation... \n",
    "(tm.train_tfmsg[-1:], tm.valid_tfmsg[-1:])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Create multi spectrogram transformation:\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "89fca47b3bc84ab5a122831275dca36b",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Tab(children=(Output(), VBox(children=(Label(value='(0)\\tfunctools.partial(<function tfm_add_white_noise at 0x…"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "s = sl[0] #sample without label\n",
    "sAug = tm.try_data_augmentation(s) #apply data augmentation and show results..."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Create multi channel transform"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [],
   "source": [
    "def tfm_trim(ad:AudioData, xBegin, xEnd):\n",
    "    '''Extract part of original signal'''\n",
    "    #print(ad.sig.shape, xBegin, xEnd)\n",
    "    x = ad.sig[int(xBegin):int(xEnd)]\n",
    "    return AudioData(x,ad.sr)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "DONE\n"
     ]
    }
   ],
   "source": [
    "def test_tfm_trim():\n",
    "    assert 100==tfm_trim(s,0,100).sig.shape[0]\n",
    "    assert 100==tfm_trim(s,20,20+100).sig.shape[0]\n",
    "    print('DONE')\n",
    "test_tfm_trim()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "10000 0 1600.0 3200.0\n"
     ]
    }
   ],
   "source": [
    "#Parameters\n",
    "wndWidth =  6800\n",
    "n_windows = 3\n",
    "wndStep = (max_duration - wndWidth)/(n_windows-1)\n",
    "start1 = 0\n",
    "start2 = wndStep*1\n",
    "start3 = wndStep*2\n",
    "\n",
    "print(max_duration, start1,start2,start3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [],
   "source": [
    "def slidingSpectrogram(ad:AudioData, wndStart=0, wndWidth=6800):\n",
    "    trimmed = tfm_trim(ad, wndStart, wndStart+wndWidth) # Trim to the appropriate size\n",
    "    minLenForSpec = 6401\n",
    "    maxWW = max(minLenForSpec,wndWidth) # Cant be shorter...\n",
    "    padded = tfm_pad_to_max(trimmed, maxWW) # Ensure constant lenght\n",
    "    spec = tfm_spectro(padded, n_mels=224, ws=50, n_fft=3200*2, to_db_scale=True) # Spectrogram\n",
    "    #print(spec.shape)\n",
    "    ret =  spec[:,:,:120] # Crop\n",
    "    #print(ret.shape)\n",
    "    return ret"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [],
   "source": [
    "def tfm_cropSpec(spec:torch.Tensor, yStart, yEnd)->torch.Tensor:\n",
    "    #yStart = int(math.floor(yStart))\n",
    "    #yEnd = int(math.floor(yEnd))\n",
    "    #print(spec.shape, yStart, yEnd, yEnd-yStart)\n",
    "    ret = spec[:,:,yStart:yEnd]\n",
    "    #print(ret.shape)\n",
    "    return ret"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [],
   "source": [
    "def tfm_multiSpectrumSlide(ad:AudioData)->torch.Tensor:\n",
    "    '''This transformation is going to produce a 3-channels spectrum image from sliding windows'''\n",
    "    minLenForSpec = 6401\n",
    "    maxWW = max(minLenForSpec,max_duration) # max width\n",
    "    \n",
    "    padded = tfm_pad_to_max(ad, maxWW) # Ensure constant lenght\n",
    "    spec = tfm_spectro(padded, n_mels=224, ws=50, n_fft=3200*2, to_db_scale=True) # Spectrogram of all\n",
    "\n",
    "    wndWidth = 140 # in spec cols\n",
    "    maxSpecCols = spec.shape[2]\n",
    "    n_windows = 3\n",
    "    wndStep = (maxSpecCols - wndWidth)/(n_windows-1)\n",
    "    wndStep = int(math.floor(wndStep)) # make it integer\n",
    "    \n",
    "    #print(wndWidth, wndStep)\n",
    "    \n",
    "    n=0\n",
    "    ret = None\n",
    "    ret = tfm_cropSpec(spec,(n*wndStep),((n)*wndStep)+wndWidth) # Sliding window\n",
    "    wMax = ret.shape[2]\n",
    "    \n",
    "    n=1\n",
    "    img = tfm_cropSpec(spec,(n*wndStep),((n)*wndStep)+wndWidth)\n",
    "    ret = torch.cat([ret,img],dim=0) # Stack...\n",
    "\n",
    "    n=2\n",
    "    img = tfm_cropSpec(spec,(n*wndStep),((n)*wndStep)+wndWidth)\n",
    "    ret = torch.cat([ret,img],dim=0) # Stack...\n",
    "\n",
    "    return ret"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/jpeg": "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\n",
      "image/png": "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\n",
      "text/plain": [
       "Image (1, 224, 140)"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/jpeg": "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\n",
      "image/png": "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\n",
      "text/plain": [
       "Image (1, 224, 140)"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/jpeg": "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\n",
      "image/png": "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\n",
      "text/plain": [
       "Image (1, 224, 140)"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "#Display channels inepenently\n",
    "imgMc = tfm_multiSpectrumSlide(sAug)\n",
    "for i in range(imgMc.shape[0]):\n",
    "    display(Image(imgMc[i:i+1,:,:]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "3e38387a0e054db0a5ff9f4f9dc84cba",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "VBox(children=(HTML(value='<div style=\"background-color:black;color:white;text-align:center;\">None</div>', lay…"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "None"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "tm.try_train_tfms(s);"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor([[[2.6595e-02, 2.8827e-02, 2.8901e-02,  ..., 0.0000e+00,\n",
       "          0.0000e+00, 0.0000e+00],\n",
       "         [1.9920e-01, 2.5152e-01, 2.9225e-01,  ..., 0.0000e+00,\n",
       "          0.0000e+00, 0.0000e+00],\n",
       "         [1.0051e-01, 1.3693e-01, 1.5372e-01,  ..., 0.0000e+00,\n",
       "          0.0000e+00, 0.0000e+00],\n",
       "         ...,\n",
       "         [1.1172e-05, 5.1642e-05, 4.9683e-04,  ..., 0.0000e+00,\n",
       "          0.0000e+00, 0.0000e+00],\n",
       "         [6.6122e-05, 1.5709e-05, 5.3219e-04,  ..., 0.0000e+00,\n",
       "          0.0000e+00, 0.0000e+00],\n",
       "         [1.1435e-04, 2.1504e-05, 2.4337e-04,  ..., 0.0000e+00,\n",
       "          0.0000e+00, 0.0000e+00]]])"
      ]
     },
     "execution_count": 47,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tm.apply_valid_tfms(s)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Create final transformations\n",
    "\n",
    "We now create the final transofrmation that outputs a \"multi-channel\" image where channels:\n",
    " - 0: 100% signal\n",
    " - 1: 50% gignal made doubling the X resolution of spectrogram \n",
    " - 2: 25% gignal made *4 the X resolution of spectrogram\n",
    " \n",
    "The expected result on the final training is to improve classification performance on \"short\" sounds that are actually the most common.\n",
    "\n",
    "**IMPORTANT**: now our output tensor is a 3-channel tensor so we can use the standard architecture\n",
    " \n",
    "**TODO** : possible improvements\n",
    " - Increaze the number of zooms\n",
    " - Sliding window to cope with long signals\n",
    " - Stack multiple different transformations at one"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "([<function __main__.tfm_multiSpectrumSlide(ad: exp.nb_AudioCommon.AudioData) -> torch.Tensor>],\n",
       " [<function __main__.tfm_multiSpectrumSlide(ad: exp.nb_AudioCommon.AudioData) -> torch.Tensor>])"
      ]
     },
     "execution_count": 48,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#Replace the spectrogram transform with tfm_spectro_stft\n",
    "tm.train_tfmsg[-1]=[tfm_multiSpectrumSlide]\n",
    "tm.valid_tfmsg[-1]=[tfm_multiSpectrumSlide]\n",
    "(tm.train_tfmsg[-1], tm.valid_tfmsg[-1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "bc916a8aec734bd8850515b4ea64cb85",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "VBox(children=(HTML(value='<div style=\"background-color:black;color:white;text-align:center;\">None</div>', lay…"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "None"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "#Try single one\n",
    "tm.try_train_tfms(s);"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### DataBlock"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(61, ['aa', 'ae', 'ah'])"
      ]
     },
     "execution_count": 49,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#Create classes for all data!\n",
    "classes = list(sorted(all_phonemes_df.PHN.unique()))\n",
    "len(classes), classes[:3]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[[functools.partial(<function tfm_add_white_noise at 0x7f4cb8cbf9d8>, noise_scl=0.0001, lower_gain=0.9, upper_gain=1.05),\n",
       "  functools.partial(<function tfm_modulate_volume at 0x7f4cb8cbfa60>, noise_scl=0.0001, lower_gain=0.9, upper_gain=1.05),\n",
       "  functools.partial(<function tfm_pad_to_max at 0x7f4cb8cbfd90>, mx=10000),\n",
       "  <function __main__.tfm_multiSpectrumSlide(ad: exp.nb_AudioCommon.AudioData) -> torch.Tensor>],\n",
       " [functools.partial(<function tfm_pad_to_max at 0x7f4cb8cbfd90>, mx=10000),\n",
       "  <function __main__.tfm_multiSpectrumSlide(ad: exp.nb_AudioCommon.AudioData) -> torch.Tensor>]]"
      ]
     },
     "execution_count": 50,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tfms = tm.get_tfms()\n",
    "tfms"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [],
   "source": [
    "data = (AudioList.from_df(src, path, cols=['SampleAndSr'])\n",
    "        .split_from_df('is_valid')\n",
    "        .label_from_df('PHN', classes=classes)\n",
    "        .transform(tfms)\n",
    "        .databunch(bs=64)) # NOTE: "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Learner"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn = cnn_learner(data, models.resnet50, metrics=accuracy)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**No need to adapt the first layer: is actually a 3-channels image**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "First layer shape: torch.Size([64, 3, 7, 7])\n"
     ]
    }
   ],
   "source": [
    "#adapt_first_layer(learn.model,nChannels)\n",
    "print(f'First layer shape: {learn.model[0][0].weight.shape}')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Train"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LR Finder is complete, type {learner_name}.recorder.plot() to see the graph.\n",
      "Min numerical gradient: 5.25E-03\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "lr_find(learn)\n",
    "learn.recorder.plot(suggestion=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "\n",
       "    <div>\n",
       "        <style>\n",
       "            /* Turns off some styling */\n",
       "            progress {\n",
       "                /* gets rid of default border in Firefox and Opera. */\n",
       "                border: none;\n",
       "                /* Needs to be in here for Safari polyfill so background images work as expected. */\n",
       "                background-size: auto;\n",
       "            }\n",
       "            .progress-bar-interrupted, .progress-bar-interrupted::-webkit-progress-bar {\n",
       "                background: #F44336;\n",
       "            }\n",
       "        </style>\n",
       "      <progress value='5' class='' max='8', style='width:300px; height:20px; vertical-align: middle;'></progress>\n",
       "      62.50% [5/8 2:19:14<1:23:32]\n",
       "    </div>\n",
       "    \n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: left;\">\n",
       "      <th>epoch</th>\n",
       "      <th>train_loss</th>\n",
       "      <th>valid_loss</th>\n",
       "      <th>accuracy</th>\n",
       "      <th>time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>1.528819</td>\n",
       "      <td>1.456701</td>\n",
       "      <td>0.534554</td>\n",
       "      <td>27:45</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>1.355649</td>\n",
       "      <td>1.231745</td>\n",
       "      <td>0.608129</td>\n",
       "      <td>27:47</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>1.262017</td>\n",
       "      <td>1.180571</td>\n",
       "      <td>0.626319</td>\n",
       "      <td>27:52</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>1.177578</td>\n",
       "      <td>1.092531</td>\n",
       "      <td>0.649991</td>\n",
       "      <td>27:52</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>1.056059</td>\n",
       "      <td>1.004916</td>\n",
       "      <td>0.672631</td>\n",
       "      <td>27:55</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table><p>\n",
       "\n",
       "    <div>\n",
       "        <style>\n",
       "            /* Turns off some styling */\n",
       "            progress {\n",
       "                /* gets rid of default border in Firefox and Opera. */\n",
       "                border: none;\n",
       "                /* Needs to be in here for Safari polyfill so background images work as expected. */\n",
       "                background-size: auto;\n",
       "            }\n",
       "            .progress-bar-interrupted, .progress-bar-interrupted::-webkit-progress-bar {\n",
       "                background: #F44336;\n",
       "            }\n",
       "        </style>\n",
       "      <progress value='67' class='' max='1001', style='width:300px; height:20px; vertical-align: middle;'></progress>\n",
       "      6.69% [67/1001 00:31<07:17]\n",
       "    </div>\n",
       "    "
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "IOPub message rate exceeded.\n",
      "The notebook server will temporarily stop sending output\n",
      "to the client in order to avoid crashing it.\n",
      "To change this limit, set the config variable\n",
      "`--NotebookApp.iopub_msg_rate_limit`.\n",
      "\n",
      "Current values:\n",
      "NotebookApp.iopub_msg_rate_limit=1000.0 (msgs/sec)\n",
      "NotebookApp.rate_limit_window=3.0 (secs)\n",
      "\n"
     ]
    }
   ],
   "source": [
    "learn.fit_one_cycle(8, slice(1e-2))\n",
    "learn.recorder.plot_losses()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "learn.recorder.plot_losses()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Impressive results with all data: algorithm generalizes very very well!"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "Total time: 22:29 <p><table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: left;\">\n",
       "      <th>epoch</th>\n",
       "      <th>train_loss</th>\n",
       "      <th>valid_loss</th>\n",
       "      <th>accuracy</th>\n",
       "      <th>time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>2.486425</td>\n",
       "      <td>2.214132</td>\n",
       "      <td>0.337423</td>\n",
       "      <td>02:48</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>2.066756</td>\n",
       "      <td>1.899197</td>\n",
       "      <td>0.409313</td>\n",
       "      <td>02:47</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>1.758308</td>\n",
       "      <td>1.635987</td>\n",
       "      <td>0.487651</td>\n",
       "      <td>02:49</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>1.547204</td>\n",
       "      <td>1.521214</td>\n",
       "      <td>0.524304</td>\n",
       "      <td>02:47</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>1.375261</td>\n",
       "      <td>1.379541</td>\n",
       "      <td>0.560327</td>\n",
       "      <td>02:49</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>5</td>\n",
       "      <td>1.235152</td>\n",
       "      <td>1.331511</td>\n",
       "      <td>0.575114</td>\n",
       "      <td>02:48</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>6</td>\n",
       "      <td>1.020621</td>\n",
       "      <td>1.305962</td>\n",
       "      <td>0.590530</td>\n",
       "      <td>02:48</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>7</td>\n",
       "      <td>0.947746</td>\n",
       "      <td>1.308162</td>\n",
       "      <td>0.594306</td>\n",
       "      <td>02:48</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "learn.fit_one_cycle(8, slice(1e-2))\n",
    "learn.recorder.plot_losses()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**NB:** sliding window is about 40% faster than multi-resolution"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.save('pass1_all_mc')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.freeze_to(1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LR Finder is complete, type {learner_name}.recorder.plot() to see the graph.\n",
      "Min numerical gradient: 7.59E-07\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "lr_find(learn)\n",
    "learn.recorder.plot(suggestion=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "\n",
       "    <div>\n",
       "        <style>\n",
       "            /* Turns off some styling */\n",
       "            progress {\n",
       "                /* gets rid of default border in Firefox and Opera. */\n",
       "                border: none;\n",
       "                /* Needs to be in here for Safari polyfill so background images work as expected. */\n",
       "                background-size: auto;\n",
       "            }\n",
       "            .progress-bar-interrupted, .progress-bar-interrupted::-webkit-progress-bar {\n",
       "                background: #F44336;\n",
       "            }\n",
       "        </style>\n",
       "      <progress value='6' class='' max='8', style='width:300px; height:20px; vertical-align: middle;'></progress>\n",
       "      75.00% [6/8 2:47:04<55:41]\n",
       "    </div>\n",
       "    \n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: left;\">\n",
       "      <th>epoch</th>\n",
       "      <th>train_loss</th>\n",
       "      <th>valid_loss</th>\n",
       "      <th>accuracy</th>\n",
       "      <th>time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>0.865493</td>\n",
       "      <td>0.909029</td>\n",
       "      <td>0.700753</td>\n",
       "      <td>27:49</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>0.865160</td>\n",
       "      <td>0.911118</td>\n",
       "      <td>0.700206</td>\n",
       "      <td>27:51</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>0.834766</td>\n",
       "      <td>0.908871</td>\n",
       "      <td>0.702455</td>\n",
       "      <td>27:50</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>0.773275</td>\n",
       "      <td>0.884291</td>\n",
       "      <td>0.709309</td>\n",
       "      <td>27:47</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>0.725714</td>\n",
       "      <td>0.883486</td>\n",
       "      <td>0.713135</td>\n",
       "      <td>27:52</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>5</td>\n",
       "      <td>0.679396</td>\n",
       "      <td>0.882158</td>\n",
       "      <td>0.715211</td>\n",
       "      <td>27:53</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table><p>\n",
       "\n",
       "    <div>\n",
       "        <style>\n",
       "            /* Turns off some styling */\n",
       "            progress {\n",
       "                /* gets rid of default border in Firefox and Opera. */\n",
       "                border: none;\n",
       "                /* Needs to be in here for Safari polyfill so background images work as expected. */\n",
       "                background-size: auto;\n",
       "            }\n",
       "            .progress-bar-interrupted, .progress-bar-interrupted::-webkit-progress-bar {\n",
       "                background: #F44336;\n",
       "            }\n",
       "        </style>\n",
       "      <progress value='500' class='' max='2762', style='width:300px; height:20px; vertical-align: middle;'></progress>\n",
       "      18.10% [500/2762 03:43<16:51 0.6029]\n",
       "    </div>\n",
       "    "
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "IOPub message rate exceeded.\n",
      "The notebook server will temporarily stop sending output\n",
      "to the client in order to avoid crashing it.\n",
      "To change this limit, set the config variable\n",
      "`--NotebookApp.iopub_msg_rate_limit`.\n",
      "\n",
      "Current values:\n",
      "NotebookApp.iopub_msg_rate_limit=1000.0 (msgs/sec)\n",
      "NotebookApp.rate_limit_window=3.0 (secs)\n",
      "\n",
      "IOPub message rate exceeded.\n",
      "The notebook server will temporarily stop sending output\n",
      "to the client in order to avoid crashing it.\n",
      "To change this limit, set the config variable\n",
      "`--NotebookApp.iopub_msg_rate_limit`.\n",
      "\n",
      "Current values:\n",
      "NotebookApp.iopub_msg_rate_limit=1000.0 (msgs/sec)\n",
      "NotebookApp.rate_limit_window=3.0 (secs)\n",
      "\n"
     ]
    }
   ],
   "source": [
    "learn.fit_one_cycle(8, slice(1e-4))\n",
    "learn.recorder.plot_losses()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "learn.recorder.plot_losses()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.save('pass2_all')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.unfreeze()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LR Finder is complete, type {learner_name}.recorder.plot() to see the graph.\n",
      "Min numerical gradient: 1.91E-06\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "lr_find(learn)\n",
    "learn.recorder.plot(suggestion=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "Total time: 3:46:50 <p><table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: left;\">\n",
       "      <th>epoch</th>\n",
       "      <th>train_loss</th>\n",
       "      <th>valid_loss</th>\n",
       "      <th>accuracy</th>\n",
       "      <th>time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>0.585950</td>\n",
       "      <td>0.893180</td>\n",
       "      <td>0.715789</td>\n",
       "      <td>27:51</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>0.571960</td>\n",
       "      <td>0.907125</td>\n",
       "      <td>0.712729</td>\n",
       "      <td>27:57</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>0.573770</td>\n",
       "      <td>0.918177</td>\n",
       "      <td>0.711761</td>\n",
       "      <td>27:55</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>0.550138</td>\n",
       "      <td>0.923194</td>\n",
       "      <td>0.713353</td>\n",
       "      <td>27:59</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>0.517276</td>\n",
       "      <td>0.935085</td>\n",
       "      <td>0.711605</td>\n",
       "      <td>29:10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>5</td>\n",
       "      <td>0.474459</td>\n",
       "      <td>0.941725</td>\n",
       "      <td>0.710917</td>\n",
       "      <td>29:33</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>6</td>\n",
       "      <td>0.464469</td>\n",
       "      <td>0.944859</td>\n",
       "      <td>0.710777</td>\n",
       "      <td>28:12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>7</td>\n",
       "      <td>0.453244</td>\n",
       "      <td>0.945796</td>\n",
       "      <td>0.711464</td>\n",
       "      <td>28:09</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "learn.fit_one_cycle(8, slice(1e-6, 1.1e-5))\n",
    "learn.recorder.plot_losses()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Not improving..."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "metadata": {},
   "outputs": [],
   "source": [
    "#learn.save('pass3_all')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Impressive about 40% better with all data compared to 10%."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Fin."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.0"
  },
  "varInspector": {
   "cols": {
    "lenName": 16,
    "lenType": 16,
    "lenVar": 40
   },
   "kernels_config": {
    "python": {
     "delete_cmd_postfix": "",
     "delete_cmd_prefix": "del ",
     "library": "var_list.py",
     "varRefreshCmd": "print(var_dic_list())"
    },
    "r": {
     "delete_cmd_postfix": ") ",
     "delete_cmd_prefix": "rm(",
     "library": "var_list.r",
     "varRefreshCmd": "cat(var_dic_list()) "
    }
   },
   "types_to_exclude": [
    "module",
    "function",
    "builtin_function_or_method",
    "instance",
    "_Feature"
   ],
   "window_display": false
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}