wavenet.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "%matplotlib inline\n",
    "import matplotlib.pyplot as plt      \n",
    "import numpy as np\n",
    "import torch\n",
    "from torch import nn\n",
    "import torch.nn.functional as F\n",
    "from torch.autograd import Variable\n",
    "from torch.utils.data import DataLoader\n",
    "from torch.optim import Adam\n",
    "from datasets import WaveNetDataset"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Sine Wave to test with Wavenet"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "PATH = '/home/austin/data/tacotron/LJSpeech-1.0'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "dataset = WaveNetDataset(PATH)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "s, a = dataset[1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(153, 80)"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "s.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(41885,)"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "a.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "256"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.prod([4, 4, 4, 4])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "273.75816993464053"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(a) / s.shape[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "sine_wave = np.sin(np.arange(0, 10, 0.1))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAZQAAAEKCAYAAAA1qaOTAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fac727916a0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(sine_wave)\n",
    "plt.xlabel('Angle [rad]')\n",
    "plt.ylabel('sin(x)')\n",
    "plt.axis('tight');\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Encoding from the first wavenet paper, tacotron2 uses MoL encoding instead"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "def mu_law_encode(signal, mu=256):\n",
    "    \"\"\"\n",
    "    Quantizes a signal to mu number discrete values.\n",
    "    \"\"\"\n",
    "    mu = mu-1\n",
    "    fx = np.sign(signal) * (np.log(1 + mu*np.abs(signal))/\n",
    "                           np.log(1 + mu))\n",
    "    return np.floor((fx+1)/2*mu+0.5).astype(np.long)\n",
    "\n",
    "\n",
    "def decode_mu_law(quantized_signal, mu=256):\n",
    "    mu = mu-1\n",
    "    fx = (quantized_signal-0.5)/mu*2-1\n",
    "    x = (np.sign(fx)/\n",
    "         mu*((1+mu)**np.abs(fx)-1))\n",
    "    return x\n",
    "\n",
    "\n",
    "def one_hot_encode(indices, num_categories=256):\n",
    "    batch_size, sequence_length = indices.size()\n",
    "    one_hot = torch.zeros(batch_size, num_categories, sequence_length)\n",
    "    one_hot.scatter_(1, indices.unsqueeze(dim=1), 1.0)\n",
    "    return Variable(one_hot)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "mu_sine = mu_law_encode(sine_wave)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXoAAAD8CAYAAAB5Pm/hAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fac71d10940>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(mu_sine,'.', ms=2);"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYYAAAD8CAYAAABzTgP2AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nZmbW4f8BnohJw+KfJBIAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fac71c7f470>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(decode_mu_law(mu_sine), '.', ms=2);"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "class OneHot(nn.Module):\n",
    "    \"\"\"\n",
    "    One hot encodes a categorical sequence\n",
    "    functions similar to an embedding layer\n",
    "    \"\"\"\n",
    "    \n",
    "    def __init__(self, num_categories=256):\n",
    "        super(OneHot, self).__init__()\n",
    "        self.num_categories = num_categories\n",
    "        \n",
    "    def forward(self, indices):\n",
    "        indices = indices.data\n",
    "        batch_size, sequence_length = indices.size()\n",
    "        one_hot = indices.new(batch_size, self.num_categories, sequence_length).zero_()\n",
    "        one_hot.scatter_(1, indices.unsqueeze(dim=1), 1)\n",
    "        return Variable(one_hot.float())   "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Causal Convolution"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "class CausalConv1d(nn.Module):\n",
    "    \"\"\"\n",
    "    pads the left side of input sequence just enough so that the convolutional kernel\n",
    "    does not look into the future.\n",
    "    \n",
    "    Input and output sizes will be the same.\n",
    "    \"\"\"\n",
    "    def __init__(self, in_channels, out_channels, kernel_size, dilation=1):\n",
    "        super(CausalConv1d, self).__init__()\n",
    "        pad = (kernel_size - 1) * dilation\n",
    "        self.conv1 = nn.Conv1d(in_channels, out_channels, kernel_size, padding=pad, dilation=dilation)\n",
    "\n",
    "    def forward(self, x):\n",
    "        x = self.conv1(x)\n",
    "        x = x[..., :-self.conv1.padding[0]]  # remove trailing padding\n",
    "        return x"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "class ResidualLayer(nn.Module):\n",
    "    \"\"\"\n",
    "    A wavenet causal gated residual layer.\n",
    "    \"\"\"\n",
    "    \n",
    "    def __init__(self, residual_channels, skip_channels, dilation, conditioning_channels=None):\n",
    "        super(ResidualLayer, self).__init__()\n",
    "        self.conv_filter = CausalConv1d(residual_channels, residual_channels,\n",
    "                                         kernel_size=3, dilation=dilation)\n",
    "        self.conv_gate = CausalConv1d(residual_channels, residual_channels,\n",
    "                                         kernel_size=3, dilation=dilation)        \n",
    "        self.resconv = nn.Conv1d(residual_channels, residual_channels, kernel_size=1)\n",
    "        self.skipconv = nn.Conv1d(residual_channels, skip_channels, kernel_size=1)\n",
    "        \n",
    "        if conditioning_channels:\n",
    "            self.conditioning_conv1x1 = nn.Conv1d(conditioning_channels, residual_channels * 2, kernel_size=1)\n",
    "\n",
    "            \n",
    "        \n",
    "    def forward(self, x, local_conditioning=None):\n",
    "        conv_filter = self.conv_filter(x)\n",
    "        conv_gate = self.conv_gate(x)\n",
    "        \n",
    "        if local_conditioning is not None:\n",
    "            local_conditioning = self.conditioning_conv1x1(local_conditioning)\n",
    "            local_conditioning_filter, local_conditioning_gate = local_conditioning.chunk(chunks=2, dim=1)\n",
    "            conv_filter += local_conditioning_filter\n",
    "            conv_gate += local_conditioning_gate\n",
    "            \n",
    "        fx = F.tanh(conv_filter) * F.sigmoid(conv_gate)\n",
    "        fx = self.resconv(fx) # restore feature dims with \n",
    "        skip = self.skipconv(fx)  # conv1x1 goes to skip connections\n",
    "        residual = fx + x  # residual output goes to next layer\n",
    "        return skip, residual\n",
    "        "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "class DilatedStack(nn.Module):\n",
    "    def __init__(self, residual_channels, skip_channels, dilation_depth, conditioning_channels=None):\n",
    "        \"\"\"\n",
    "        Block of dilated residual layers\n",
    "        Dilation increases exponentially, final dilation will be 2**num_layers.\n",
    "        a 10 layer block has a receptive field of 1024 (kernel_size * dilation).\n",
    "        \"\"\"\n",
    "        super(DilatedStack, self).__init__()\n",
    "        residual_stack = [ResidualLayer(residual_channels, skip_channels, 2**layer, conditioning_channels)\n",
    "                         for layer in range(dilation_depth)]\n",
    "        self.residual_stack = nn.ModuleList(residual_stack)\n",
    "        \n",
    "    def forward(self, x, local_conditioning=None):\n",
    "        skips = []\n",
    "        for layer in self.residual_stack:\n",
    "            skip, x = layer(x, local_conditioning)\n",
    "            skips.append(skip.unsqueeze(0))\n",
    "        return torch.cat(skips, dim=0), x  # layers, batch, features, seq\n",
    "    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "class UpsamplingNetwork(nn.Module):\n",
    "    \"\"\"Upsamples local conditioning variables to the same length as input.\n",
    "\n",
    "    \"\"\"\n",
    "    def __init__(self, conditioning_channels, scales):\n",
    "        super(UpsamplingNetwork, self).__init__()\n",
    "        upsample_layers = [nn.ConvTranspose1d(conditioning_channels, conditioning_channels,\n",
    "                                        kernel_size=3, stride=scale, padding=1, output_padding=1)\n",
    "                           for scale in scales]\n",
    "        self.upsample_layers = nn.ModuleList(upsample_layers)\n",
    "\n",
    "    def forward(self, x):\n",
    "        for upsample in self.upsample_layers:\n",
    "            x = upsample(x)\n",
    "        return x"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "class WaveNet(nn.Module):\n",
    "    def __init__(self, residual_channels, skip_channels, dilation_cycles, dilation_depth, conditioning_channels=None, upsampling_scales=None):\n",
    "        super(WaveNet, self).__init__()\n",
    "        self.conditioning_channels = conditioning_channels\n",
    "        self.upsampling_scales = upsampling_scales\n",
    "        self.one_hot_encode = OneHot(num_categories=256)\n",
    "        self.input_conv = CausalConv1d(in_channels=256, out_channels=residual_channels, kernel_size=3)        \n",
    "        self.dilated_stacks = nn.ModuleList(\n",
    "            [DilatedStack(residual_channels, skip_channels, dilation_depth, conditioning_channels)\n",
    "             for cycle in range(dilation_cycles)]\n",
    "        )\n",
    "        self.convout_1 = nn.Conv1d(in_channels=256, out_channels=256, kernel_size=1)\n",
    "        self.convout_2 = nn.Conv1d(in_channels=256, out_channels=256, kernel_size=1)\n",
    "        \n",
    "        if conditioning_channels and upsampling_scales:\n",
    "            self.local_conditioning_upsample = UpsamplingNetwork(conditioning_channels, upsampling_scales)\n",
    "\n",
    "        \n",
    "    def forward(self, x, local_conditioning=None):\n",
    "        x = self.one_hot_encode(x) # [batch, num_mu, seq_len]\n",
    "        x = self.input_conv(x) # [1, 256, 100] expects one_hot encoded inputs\n",
    "        \n",
    "        if local_conditioning is not None:\n",
    "            if not self.conditioning_channels and not self.upsampling_scales:\n",
    "                raise ValueError('set conditioning_channels, and upsampling_scales to use local_conditioning.')    \n",
    "            local_conditioning = self.local_conditioning_upsample(local_conditioning)\n",
    "            \n",
    "        skip_connections = []\n",
    "        for cycle in self.dilated_stacks:\n",
    "            skips, x = cycle(x, local_conditioning) # [5, 1, 16, 100] and [1, 32, 100]\n",
    "            skip_connections.append(skips)\n",
    "        skip_connections = torch.cat(skip_connections, dim=0) # [10, 1, 16, 100]\n",
    "        \n",
    "        # gather all output skip connections to generate output, discard last residual output\n",
    "        out = skip_connections.sum(dim=0) # [1, 16, 100]\n",
    "        out = F.relu(out)\n",
    "        out = self.convout_1(out) # [1, 256, 100]\n",
    "        out = F.relu(out)\n",
    "        return self.convout_2(out)\n",
    "    \n",
    "    def generate(self, start, local_conditioning, maxlen=100):\n",
    "        outputs = [start]\n",
    "        for i in range(maxlen):\n",
    "            probs = self.forward(torch.cat(outputs, dim=1), local_conditioning) # P(all_next|all previous)\n",
    "            _, output = probs[..., -1:].max(dim=1) # get prob for the last word and take max idx\n",
    "            outputs.append(output)\n",
    "        return torch.cat(outputs, dim=1)[..., start.size(dim=1):]\n",
    "     "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [],
   "source": [
    "def sine_generator(seq_size= 30001, mu=256):\n",
    "    framerate = 44100\n",
    "    t = np.linspace(0, 5, framerate*5)\n",
    "    data = np.sin(2*np.pi*220*t) + np.sin(2*np.pi*224*t)\n",
    "    data = data/2\n",
    "    while True:\n",
    "        start = np.random.randint(0, data.shape[0] - seq_size)\n",
    "        sine = data[start:start+seq_size]\n",
    "        mu_sine = mu_law_encode(sine, mu)\n",
    "        mu_sine = torch.from_numpy(mu_sine).unsqueeze(0)\n",
    "        yield Variable(mu_sine[..., :seq_size]).cuda()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "wavenet = WaveNet(residual_channels=512, skip_channels=256, dilation_cycles=2, dilation_depth=10, conditioning_channels=1, upsampling_scales=[2, 2])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "optimizer = Adam(wavenet.parameters(), lr=1e-2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "g = sine_generator(seq_size=3001)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "batch = next(g)\n",
    "conditional = batch[:, :750].view(1, 1, -1).float()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Variable containing:\n",
       "( 0  ,.,.) = \n",
       "  6.6366e-02  8.9705e-03 -1.3580e-01  ...   1.6499e-03 -7.1411e-02  1.0277e-01\n",
       " -8.4379e-02 -2.4416e-01 -3.2983e-01  ...  -1.7572e-01 -2.3323e-01 -1.2676e-01\n",
       "  1.5932e-02  1.1617e-02  4.2358e-02  ...  -1.6942e-02  4.9782e-02 -6.1901e-02\n",
       "                 ...                   ⋱                   ...                \n",
       "  3.2938e-02  4.9972e-03 -3.2651e-02  ...   7.1967e-02  3.5655e-02  5.6869e-02\n",
       " -4.3091e-02  1.0843e-02 -1.1654e-02  ...  -3.9647e-02 -1.8494e-02 -3.4647e-02\n",
       "  1.8127e-01  9.0762e-02  9.2551e-03  ...   9.4830e-02  4.6373e-02  2.3488e-01\n",
       "[torch.FloatTensor of size 1x256x3000]"
      ]
     },
     "execution_count": 30,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "wavenet(batch[:, :-1].cpu(), conditional.cpu())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "WaveNet(\n",
       "  (one_hot_encode): OneHot(\n",
       "  )\n",
       "  (input_conv): CausalConv1d(\n",
       "    (conv1): Conv1d(256, 512, kernel_size=(3,), stride=(1,), padding=(2,))\n",
       "  )\n",
       "  (dilated_stacks): ModuleList(\n",
       "    (0): DilatedStack(\n",
       "      (residual_stack): ModuleList(\n",
       "        (0): ResidualLayer(\n",
       "          (conv_filter): CausalConv1d(\n",
       "            (conv1): Conv1d(512, 512, kernel_size=(3,), stride=(1,), padding=(2,))\n",
       "          )\n",
       "          (conv_gate): CausalConv1d(\n",
       "            (conv1): Conv1d(512, 512, kernel_size=(3,), stride=(1,), padding=(2,))\n",
       "          )\n",
       "          (resconv): Conv1d(512, 512, kernel_size=(1,), stride=(1,))\n",
       "          (skipconv): Conv1d(512, 256, kernel_size=(1,), stride=(1,))\n",
       "          (conditioning_conv1x1): Conv1d(1, 1024, kernel_size=(1,), stride=(1,))\n",
       "        )\n",
       "        (1): ResidualLayer(\n",
       "          (conv_filter): CausalConv1d(\n",
       "            (conv1): Conv1d(512, 512, kernel_size=(3,), stride=(1,), padding=(4,), dilation=(2,))\n",
       "          )\n",
       "          (conv_gate): CausalConv1d(\n",
       "            (conv1): Conv1d(512, 512, kernel_size=(3,), stride=(1,), padding=(4,), dilation=(2,))\n",
       "          )\n",
       "          (resconv): Conv1d(512, 512, kernel_size=(1,), stride=(1,))\n",
       "          (skipconv): Conv1d(512, 256, kernel_size=(1,), stride=(1,))\n",
       "          (conditioning_conv1x1): Conv1d(1, 1024, kernel_size=(1,), stride=(1,))\n",
       "        )\n",
       "        (2): ResidualLayer(\n",
       "          (conv_filter): CausalConv1d(\n",
       "            (conv1): Conv1d(512, 512, kernel_size=(3,), stride=(1,), padding=(8,), dilation=(4,))\n",
       "          )\n",
       "          (conv_gate): CausalConv1d(\n",
       "            (conv1): Conv1d(512, 512, kernel_size=(3,), stride=(1,), padding=(8,), dilation=(4,))\n",
       "          )\n",
       "          (resconv): Conv1d(512, 512, kernel_size=(1,), stride=(1,))\n",
       "          (skipconv): Conv1d(512, 256, kernel_size=(1,), stride=(1,))\n",
       "          (conditioning_conv1x1): Conv1d(1, 1024, kernel_size=(1,), stride=(1,))\n",
       "        )\n",
       "        (3): ResidualLayer(\n",
       "          (conv_filter): CausalConv1d(\n",
       "            (conv1): Conv1d(512, 512, kernel_size=(3,), stride=(1,), padding=(16,), dilation=(8,))\n",
       "          )\n",
       "          (conv_gate): CausalConv1d(\n",
       "            (conv1): Conv1d(512, 512, kernel_size=(3,), stride=(1,), padding=(16,), dilation=(8,))\n",
       "          )\n",
       "          (resconv): Conv1d(512, 512, kernel_size=(1,), stride=(1,))\n",
       "          (skipconv): Conv1d(512, 256, kernel_size=(1,), stride=(1,))\n",
       "          (conditioning_conv1x1): Conv1d(1, 1024, kernel_size=(1,), stride=(1,))\n",
       "        )\n",
       "        (4): ResidualLayer(\n",
       "          (conv_filter): CausalConv1d(\n",
       "            (conv1): Conv1d(512, 512, kernel_size=(3,), stride=(1,), padding=(32,), dilation=(16,))\n",
       "          )\n",
       "          (conv_gate): CausalConv1d(\n",
       "            (conv1): Conv1d(512, 512, kernel_size=(3,), stride=(1,), padding=(32,), dilation=(16,))\n",
       "          )\n",
       "          (resconv): Conv1d(512, 512, kernel_size=(1,), stride=(1,))\n",
       "          (skipconv): Conv1d(512, 256, kernel_size=(1,), stride=(1,))\n",
       "          (conditioning_conv1x1): Conv1d(1, 1024, kernel_size=(1,), stride=(1,))\n",
       "        )\n",
       "        (5): ResidualLayer(\n",
       "          (conv_filter): CausalConv1d(\n",
       "            (conv1): Conv1d(512, 512, kernel_size=(3,), stride=(1,), padding=(64,), dilation=(32,))\n",
       "          )\n",
       "          (conv_gate): CausalConv1d(\n",
       "            (conv1): Conv1d(512, 512, kernel_size=(3,), stride=(1,), padding=(64,), dilation=(32,))\n",
       "          )\n",
       "          (resconv): Conv1d(512, 512, kernel_size=(1,), stride=(1,))\n",
       "          (skipconv): Conv1d(512, 256, kernel_size=(1,), stride=(1,))\n",
       "          (conditioning_conv1x1): Conv1d(1, 1024, kernel_size=(1,), stride=(1,))\n",
       "        )\n",
       "        (6): ResidualLayer(\n",
       "          (conv_filter): CausalConv1d(\n",
       "            (conv1): Conv1d(512, 512, kernel_size=(3,), stride=(1,), padding=(128,), dilation=(64,))\n",
       "          )\n",
       "          (conv_gate): CausalConv1d(\n",
       "            (conv1): Conv1d(512, 512, kernel_size=(3,), stride=(1,), padding=(128,), dilation=(64,))\n",
       "          )\n",
       "          (resconv): Conv1d(512, 512, kernel_size=(1,), stride=(1,))\n",
       "          (skipconv): Conv1d(512, 256, kernel_size=(1,), stride=(1,))\n",
       "          (conditioning_conv1x1): Conv1d(1, 1024, kernel_size=(1,), stride=(1,))\n",
       "        )\n",
       "        (7): ResidualLayer(\n",
       "          (conv_filter): CausalConv1d(\n",
       "            (conv1): Conv1d(512, 512, kernel_size=(3,), stride=(1,), padding=(256,), dilation=(128,))\n",
       "          )\n",
       "          (conv_gate): CausalConv1d(\n",
       "            (conv1): Conv1d(512, 512, kernel_size=(3,), stride=(1,), padding=(256,), dilation=(128,))\n",
       "          )\n",
       "          (resconv): Conv1d(512, 512, kernel_size=(1,), stride=(1,))\n",
       "          (skipconv): Conv1d(512, 256, kernel_size=(1,), stride=(1,))\n",
       "          (conditioning_conv1x1): Conv1d(1, 1024, kernel_size=(1,), stride=(1,))\n",
       "        )\n",
       "        (8): ResidualLayer(\n",
       "          (conv_filter): CausalConv1d(\n",
       "            (conv1): Conv1d(512, 512, kernel_size=(3,), stride=(1,), padding=(512,), dilation=(256,))\n",
       "          )\n",
       "          (conv_gate): CausalConv1d(\n",
       "            (conv1): Conv1d(512, 512, kernel_size=(3,), stride=(1,), padding=(512,), dilation=(256,))\n",
       "          )\n",
       "          (resconv): Conv1d(512, 512, kernel_size=(1,), stride=(1,))\n",
       "          (skipconv): Conv1d(512, 256, kernel_size=(1,), stride=(1,))\n",
       "          (conditioning_conv1x1): Conv1d(1, 1024, kernel_size=(1,), stride=(1,))\n",
       "        )\n",
       "        (9): ResidualLayer(\n",
       "          (conv_filter): CausalConv1d(\n",
       "            (conv1): Conv1d(512, 512, kernel_size=(3,), stride=(1,), padding=(1024,), dilation=(512,))\n",
       "          )\n",
       "          (conv_gate): CausalConv1d(\n",
       "            (conv1): Conv1d(512, 512, kernel_size=(3,), stride=(1,), padding=(1024,), dilation=(512,))\n",
       "          )\n",
       "          (resconv): Conv1d(512, 512, kernel_size=(1,), stride=(1,))\n",
       "          (skipconv): Conv1d(512, 256, kernel_size=(1,), stride=(1,))\n",
       "          (conditioning_conv1x1): Conv1d(1, 1024, kernel_size=(1,), stride=(1,))\n",
       "        )\n",
       "      )\n",
       "    )\n",
       "    (1): DilatedStack(\n",
       "      (residual_stack): ModuleList(\n",
       "        (0): ResidualLayer(\n",
       "          (conv_filter): CausalConv1d(\n",
       "            (conv1): Conv1d(512, 512, kernel_size=(3,), stride=(1,), padding=(2,))\n",
       "          )\n",
       "          (conv_gate): CausalConv1d(\n",
       "            (conv1): Conv1d(512, 512, kernel_size=(3,), stride=(1,), padding=(2,))\n",
       "          )\n",
       "          (resconv): Conv1d(512, 512, kernel_size=(1,), stride=(1,))\n",
       "          (skipconv): Conv1d(512, 256, kernel_size=(1,), stride=(1,))\n",
       "          (conditioning_conv1x1): Conv1d(1, 1024, kernel_size=(1,), stride=(1,))\n",
       "        )\n",
       "        (1): ResidualLayer(\n",
       "          (conv_filter): CausalConv1d(\n",
       "            (conv1): Conv1d(512, 512, kernel_size=(3,), stride=(1,), padding=(4,), dilation=(2,))\n",
       "          )\n",
       "          (conv_gate): CausalConv1d(\n",
       "            (conv1): Conv1d(512, 512, kernel_size=(3,), stride=(1,), padding=(4,), dilation=(2,))\n",
       "          )\n",
       "          (resconv): Conv1d(512, 512, kernel_size=(1,), stride=(1,))\n",
       "          (skipconv): Conv1d(512, 256, kernel_size=(1,), stride=(1,))\n",
       "          (conditioning_conv1x1): Conv1d(1, 1024, kernel_size=(1,), stride=(1,))\n",
       "        )\n",
       "        (2): ResidualLayer(\n",
       "          (conv_filter): CausalConv1d(\n",
       "            (conv1): Conv1d(512, 512, kernel_size=(3,), stride=(1,), padding=(8,), dilation=(4,))\n",
       "          )\n",
       "          (conv_gate): CausalConv1d(\n",
       "            (conv1): Conv1d(512, 512, kernel_size=(3,), stride=(1,), padding=(8,), dilation=(4,))\n",
       "          )\n",
       "          (resconv): Conv1d(512, 512, kernel_size=(1,), stride=(1,))\n",
       "          (skipconv): Conv1d(512, 256, kernel_size=(1,), stride=(1,))\n",
       "          (conditioning_conv1x1): Conv1d(1, 1024, kernel_size=(1,), stride=(1,))\n",
       "        )\n",
       "        (3): ResidualLayer(\n",
       "          (conv_filter): CausalConv1d(\n",
       "            (conv1): Conv1d(512, 512, kernel_size=(3,), stride=(1,), padding=(16,), dilation=(8,))\n",
       "          )\n",
       "          (conv_gate): CausalConv1d(\n",
       "            (conv1): Conv1d(512, 512, kernel_size=(3,), stride=(1,), padding=(16,), dilation=(8,))\n",
       "          )\n",
       "          (resconv): Conv1d(512, 512, kernel_size=(1,), stride=(1,))\n",
       "          (skipconv): Conv1d(512, 256, kernel_size=(1,), stride=(1,))\n",
       "          (conditioning_conv1x1): Conv1d(1, 1024, kernel_size=(1,), stride=(1,))\n",
       "        )\n",
       "        (4): ResidualLayer(\n",
       "          (conv_filter): CausalConv1d(\n",
       "            (conv1): Conv1d(512, 512, kernel_size=(3,), stride=(1,), padding=(32,), dilation=(16,))\n",
       "          )\n",
       "          (conv_gate): CausalConv1d(\n",
       "            (conv1): Conv1d(512, 512, kernel_size=(3,), stride=(1,), padding=(32,), dilation=(16,))\n",
       "          )\n",
       "          (resconv): Conv1d(512, 512, kernel_size=(1,), stride=(1,))\n",
       "          (skipconv): Conv1d(512, 256, kernel_size=(1,), stride=(1,))\n",
       "          (conditioning_conv1x1): Conv1d(1, 1024, kernel_size=(1,), stride=(1,))\n",
       "        )\n",
       "        (5): ResidualLayer(\n",
       "          (conv_filter): CausalConv1d(\n",
       "            (conv1): Conv1d(512, 512, kernel_size=(3,), stride=(1,), padding=(64,), dilation=(32,))\n",
       "          )\n",
       "          (conv_gate): CausalConv1d(\n",
       "            (conv1): Conv1d(512, 512, kernel_size=(3,), stride=(1,), padding=(64,), dilation=(32,))\n",
       "          )\n",
       "          (resconv): Conv1d(512, 512, kernel_size=(1,), stride=(1,))\n",
       "          (skipconv): Conv1d(512, 256, kernel_size=(1,), stride=(1,))\n",
       "          (conditioning_conv1x1): Conv1d(1, 1024, kernel_size=(1,), stride=(1,))\n",
       "        )\n",
       "        (6): ResidualLayer(\n",
       "          (conv_filter): CausalConv1d(\n",
       "            (conv1): Conv1d(512, 512, kernel_size=(3,), stride=(1,), padding=(128,), dilation=(64,))\n",
       "          )\n",
       "          (conv_gate): CausalConv1d(\n",
       "            (conv1): Conv1d(512, 512, kernel_size=(3,), stride=(1,), padding=(128,), dilation=(64,))\n",
       "          )\n",
       "          (resconv): Conv1d(512, 512, kernel_size=(1,), stride=(1,))\n",
       "          (skipconv): Conv1d(512, 256, kernel_size=(1,), stride=(1,))\n",
       "          (conditioning_conv1x1): Conv1d(1, 1024, kernel_size=(1,), stride=(1,))\n",
       "        )\n",
       "        (7): ResidualLayer(\n",
       "          (conv_filter): CausalConv1d(\n",
       "            (conv1): Conv1d(512, 512, kernel_size=(3,), stride=(1,), padding=(256,), dilation=(128,))\n",
       "          )\n",
       "          (conv_gate): CausalConv1d(\n",
       "            (conv1): Conv1d(512, 512, kernel_size=(3,), stride=(1,), padding=(256,), dilation=(128,))\n",
       "          )\n",
       "          (resconv): Conv1d(512, 512, kernel_size=(1,), stride=(1,))\n",
       "          (skipconv): Conv1d(512, 256, kernel_size=(1,), stride=(1,))\n",
       "          (conditioning_conv1x1): Conv1d(1, 1024, kernel_size=(1,), stride=(1,))\n",
       "        )\n",
       "        (8): ResidualLayer(\n",
       "          (conv_filter): CausalConv1d(\n",
       "            (conv1): Conv1d(512, 512, kernel_size=(3,), stride=(1,), padding=(512,), dilation=(256,))\n",
       "          )\n",
       "          (conv_gate): CausalConv1d(\n",
       "            (conv1): Conv1d(512, 512, kernel_size=(3,), stride=(1,), padding=(512,), dilation=(256,))\n",
       "          )\n",
       "          (resconv): Conv1d(512, 512, kernel_size=(1,), stride=(1,))\n",
       "          (skipconv): Conv1d(512, 256, kernel_size=(1,), stride=(1,))\n",
       "          (conditioning_conv1x1): Conv1d(1, 1024, kernel_size=(1,), stride=(1,))\n",
       "        )\n",
       "        (9): ResidualLayer(\n",
       "          (conv_filter): CausalConv1d(\n",
       "            (conv1): Conv1d(512, 512, kernel_size=(3,), stride=(1,), padding=(1024,), dilation=(512,))\n",
       "          )\n",
       "          (conv_gate): CausalConv1d(\n",
       "            (conv1): Conv1d(512, 512, kernel_size=(3,), stride=(1,), padding=(1024,), dilation=(512,))\n",
       "          )\n",
       "          (resconv): Conv1d(512, 512, kernel_size=(1,), stride=(1,))\n",
       "          (skipconv): Conv1d(512, 256, kernel_size=(1,), stride=(1,))\n",
       "          (conditioning_conv1x1): Conv1d(1, 1024, kernel_size=(1,), stride=(1,))\n",
       "        )\n",
       "      )\n",
       "    )\n",
       "  )\n",
       "  (convout_1): Conv1d(256, 256, kernel_size=(1,), stride=(1,))\n",
       "  (convout_2): Conv1d(256, 256, kernel_size=(1,), stride=(1,))\n",
       "  (local_conditioning_upsample): UpsamplingNetwork(\n",
       "    (upsample_layers): ModuleList(\n",
       "      (0): ConvTranspose1d(1, 1, kernel_size=(3,), stride=(2,), padding=(1,), output_padding=(1,))\n",
       "      (1): ConvTranspose1d(1, 1, kernel_size=(3,), stride=(2,), padding=(1,), output_padding=(1,))\n",
       "    )\n",
       "  )\n",
       ")"
      ]
     },
     "execution_count": 31,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "wavenet.cuda(device=0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0 5.54139518737793\n",
      "50 5.519702434539795\n"
     ]
    }
   ],
   "source": [
    "for i in range(100): \n",
    "    batch = next(g)\n",
    "    x = batch[..., :-1]\n",
    "    y = batch[..., 1:]\n",
    "    out = wavenet(x.cuda(device=0), batch[:, :750].view(1, 1, -1).float().cuda(device=0))\n",
    "    loss = F.cross_entropy(out, y)\n",
    "    optimizer.zero_grad()\n",
    "    loss.backward()\n",
    "    optimizer.step()\n",
    "    if i % 50 == 0:\n",
    "        print(i, loss.data[0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[<matplotlib.lines.Line2D at 0x7fac55595080>]"
      ]
     },
     "execution_count": 34,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXoAAAD8CAYAAAB5Pm/hAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nEnTFAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fac69346be0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(next(g).data[0])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Exponential Time Generation zzz..."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "%time gen_wave = wavenet.generate(x[:,-1:],  batch[:, :750].view(1, 1, -1).float().cuda(device=0), maxlen=100)\n",
    "plt.plot(gen_wave.data[0]);"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.1"
  },
  "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
}