onnx_inference.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 載入前處理"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "--2023-01-21 17:58:15--  https://github.com/1010code/keyword-transformer/raw/main/data/mean.txt\n",
      "Resolving github.com (github.com)... 20.27.177.113\n",
      "Connecting to github.com (github.com)|20.27.177.113|:443... connected.\n",
      "HTTP request sent, awaiting response... 302 Found\n",
      "Location: https://raw.githubusercontent.com/1010code/keyword-transformer/main/data/mean.txt [following]\n",
      "--2023-01-21 17:58:15--  https://raw.githubusercontent.com/1010code/keyword-transformer/main/data/mean.txt\n",
      "Resolving raw.githubusercontent.com (raw.githubusercontent.com)... 185.199.110.133, 185.199.108.133, 185.199.111.133, ...\n",
      "Connecting to raw.githubusercontent.com (raw.githubusercontent.com)|185.199.110.133|:443... connected.\n",
      "HTTP request sent, awaiting response... 416 Range Not Satisfiable\n",
      "\n",
      "    The file is already fully retrieved; nothing to do.\n",
      "\n",
      "--2023-01-21 17:58:16--  https://github.com/1010code/keyword-transformer/raw/main/data/std.txt\n",
      "Resolving github.com (github.com)... 20.27.177.113\n",
      "Connecting to github.com (github.com)|20.27.177.113|:443... connected.\n",
      "HTTP request sent, awaiting response... 302 Found\n",
      "Location: https://raw.githubusercontent.com/1010code/keyword-transformer/main/data/std.txt [following]\n",
      "--2023-01-21 17:58:16--  https://raw.githubusercontent.com/1010code/keyword-transformer/main/data/std.txt\n",
      "Resolving raw.githubusercontent.com (raw.githubusercontent.com)... 185.199.108.133, 185.199.109.133, 185.199.111.133, ...\n",
      "Connecting to raw.githubusercontent.com (raw.githubusercontent.com)|185.199.108.133|:443... connected.\n",
      "HTTP request sent, awaiting response... 416 Range Not Satisfiable\n",
      "\n",
      "    The file is already fully retrieved; nothing to do.\n",
      "\n",
      "--2023-01-21 17:58:17--  https://github.com/1010code/keyword-transformer/raw/main/utils/mel_features.py\n",
      "Resolving github.com (github.com)... 20.27.177.113\n",
      "Connecting to github.com (github.com)|20.27.177.113|:443... connected.\n",
      "HTTP request sent, awaiting response... 302 Found\n",
      "Location: https://raw.githubusercontent.com/1010code/keyword-transformer/main/utils/mel_features.py [following]\n",
      "--2023-01-21 17:58:17--  https://raw.githubusercontent.com/1010code/keyword-transformer/main/utils/mel_features.py\n",
      "Resolving raw.githubusercontent.com (raw.githubusercontent.com)... 185.199.108.133, 185.199.110.133, 185.199.109.133, ...\n",
      "Connecting to raw.githubusercontent.com (raw.githubusercontent.com)|185.199.108.133|:443... connected.\n",
      "HTTP request sent, awaiting response... 416 Range Not Satisfiable\n",
      "\n",
      "    The file is already fully retrieved; nothing to do.\n",
      "\n"
     ]
    }
   ],
   "source": [
    "!wget -c https://github.com/1010code/keyword-transformer/raw/main/data/mean.txt -P data\n",
    "!wget -c https://github.com/1010code/keyword-transformer/raw/main/data/std.txt -P data\n",
    "!wget -c https://github.com/1010code/keyword-transformer/raw/main/utils/mel_features.py -P utils"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "\n",
    "mean = pd.read_csv(\"./data/mean.txt\", sep=\" \", header=None).values.reshape(98, 32)\n",
    "std = pd.read_csv(\"./data/std.txt\", sep=\" \", header=None).values.reshape(98, 32)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "import wave\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "import tensorflow as tf\n",
    "\n",
    "from utils import mel_features\n",
    "\n",
    "def load_wave(path):\n",
    "    spf = wave.open(path, \"r\")\n",
    "    # Extract Raw Audio from Wav File\n",
    "    signal = spf.readframes(-1)\n",
    "    signal = np.fromstring(signal, \"int16\")\n",
    "    return signal\n",
    "\n",
    "def wave_normalize(signal):\n",
    "    # scale to -1~1\n",
    "    MIN=-32768\n",
    "    MAX=32767;\n",
    "    input_speech=[]\n",
    "    for i in range(len(signal)):\n",
    "        X_std = (signal[i] - MIN)*1. / (MAX - MIN)\n",
    "        X_std = X_std * (1 - -1) + -1\n",
    "        input_speech.append(X_std)\n",
    "    return input_speech\n",
    "\n",
    "def get_spectrogram(waveform):\n",
    "    # Padding for files with less than 16000 samples\n",
    "    zero_padding = tf.zeros([16000] - tf.shape(waveform), dtype=tf.float32)\n",
    "\n",
    "    # Concatenate audio with padding so that all audio clips will be of the \n",
    "    # same length\n",
    "    waveform = tf.cast(waveform, tf.float32)\n",
    "\n",
    "    equal_length = tf.concat([waveform, zero_padding], 0)\n",
    "    equal_length = equal_length.numpy().flatten()\n",
    "    spectrogram = 30 * (\n",
    "        mel_features.log_mel_spectrogram(\n",
    "        equal_length,\n",
    "            16000,\n",
    "            log_offset=0.001,\n",
    "            window_length_secs=0.025,\n",
    "            hop_length_secs=0.010,\n",
    "            num_mel_bins=32,\n",
    "            lower_edge_hertz=60,\n",
    "            upper_edge_hertz=3800) - np.log(1e-3))\n",
    "\n",
    "    return spectrogram\n",
    "\n",
    "def plot_wave(wave):\n",
    "    plt.figure(figsize=(6,4))\n",
    "    plt.plot(wave)\n",
    "    plt.show()\n",
    "    plt.figure(figsize=(6,4))\n",
    "    plt.imshow(get_spectrogram(wave).T, cmap=\"inferno\")\n",
    "    plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 載入 ONNX 模型"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "!wget -c https://github.com/1010code/keyword-transformer/raw/main/data/tf_model.onnx -P data\n",
    "!wget -c https://github.com/1010code/keyword-transformer/raw/main/data/0-1.wav -P data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "!pip install onnxruntime==1.12.1 -q"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Compute the prediction with ONNX Runtime\n",
    "import onnxruntime as rt\n",
    "\n",
    "sess = rt.InferenceSession(\"./data/tf_model.onnx\")\n",
    "input_name = sess.get_inputs()[0].name\n",
    "label_name = sess.get_outputs()[0].name\n",
    "label_list = ['zero', 'one', 'two', 'three', 'four', 'five', 'six', 'seven', 'eight', 'nine']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "audio shape: (14861,)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/opt/conda/lib/python3.7/site-packages/ipykernel_launcher.py:12: DeprecationWarning: The binary mode of fromstring is deprecated, as it behaves surprisingly on unicode inputs. Use frombuffer instead\n",
      "  if sys.path[0] == '':\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"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Predict: zero\n"
     ]
    }
   ],
   "source": [
    "# 讀取資料\n",
    "signal = load_wave(\"./data/0-1.wav\")\n",
    "print(f'audio shape: {signal.shape}')\n",
    "# 正規化\n",
    "input_speech = wave_normalize(signal)\n",
    "# 繪製音訊\n",
    "plot_wave(input_speech)\n",
    "# 取得頻譜＆前處理\n",
    "data = get_spectrogram(input_speech)\n",
    "data -= -mean\n",
    "data /= std\n",
    "\n",
    "# inference\n",
    "pred_onx = sess.run([label_name], {input_name: np.array([data]).astype(np.float32)})[0]\n",
    "\n",
    "print(f'Predict: {label_list[pred_onx.argmax()]}')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "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": "3.7.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}