{ "cells": [ { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 声音识别项目介绍\n", "\n", "\n", "\n", "## 开发环境\n", "\n", "* TensorFlow的版本：2.0 +\n", "* keras\n", "* sklearn\n", "* librosa\n", "\n", "## 下载数据\n", "\n", "\n" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "!wget http://tianchi-competition.oss-cn-hangzhou.aliyuncs.com/531887/train_sample.zip\n", "!wget http://tianchi-competition.oss-cn-hangzhou.aliyuncs.com/531887/test_a.zip\n", "\n", "\n", "!unzip -qq train_sample.zip\n", "!\\rm train_sample.zip\n", "\n", "!unzip -qq test_a.zip\n", "!\\rm test_a.zip" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# 安装语音处理依赖\n", "!pip install librosa --user" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# 基本库\n", "\n", "import pandas as pd\n", "import numpy as np\n", "\n", "from sklearn.model_selection import train_test_split\n", "from sklearn.metrics import classification_report\n", "from sklearn.model_selection import GridSearchCV\n", "\n", "from sklearn.preprocessing import MinMaxScaler\n", "\n", "\n", "from tensorflow.keras.models import Sequential\n", "from tensorflow.keras.layers import Conv2D, Flatten, Dense, MaxPool2D, Dropout\n", "from tensorflow.keras.utils import to_categorical \n", "\n", "from sklearn.ensemble import RandomForestClassifier\n", "from sklearn.svm import SVC\n" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 数据预处理\n", "\n", "特征提取以及数据集的建立\n", "\n" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "feature = []\n", "label = []\n", "# 建立类别标签，不同类别对应不同的数字。\n", "label_dict = {'aloe': 0, 'burger': 1, 'cabbage': 2,'candied_fruits':3, 'carrots': 4, 'chips':5,\n", " 'chocolate': 6, 'drinks': 7, 'fries': 8, 'grapes': 9, 'gummies': 10, 'ice-cream':11,\n", " 'jelly': 12, 'noodles': 13, 'pickles': 14, 'pizza': 15, 'ribs': 16, 'salmon':17,\n", " 'soup': 18, 'wings': 19}\n", "label_dict_inv = {v:k for k,v in label_dict.items()}" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "from tqdm import tqdm\n", "def extract_features(parent_dir, sub_dirs, max_file=10, file_ext=\"*.wav\"):\n", " c = 0\n", " label, feature = [], []\n", " for sub_dir in sub_dirs:\n", " for fn in tqdm(glob.glob(os.path.join(parent_dir, sub_dir, file_ext))[:max_file]): # 遍历数据集的所有文件\n", " \n", " # segment_log_specgrams, segment_labels = [], []\n", " #sound_clip,sr = librosa.load(fn)\n", " #print(fn)\n", " label_name = fn.split('/')[-2]\n", " label.extend([label_dict[label_name]])\n", " X, sample_rate = librosa.load(fn,res_type='kaiser_fast')\n", " mels = np.mean(librosa.feature.melspectrogram(y=X,sr=sample_rate).T,axis=0) # 计算梅尔频谱(mel spectrogram),并把它作为特征\n", " feature.extend([mels])\n", " \n", " return [feature, label]" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# 自己更改目录\n", "parent_dir = './train_sample/'\n", "save_dir = \"./\"\n", "folds = sub_dirs = np.array(['aloe','burger','cabbage','candied_fruits',\n", " 'carrots','chips','chocolate','drinks','fries',\n", " 'grapes','gummies','ice-cream','jelly','noodles','pickles',\n", " 'pizza','ribs','salmon','soup','wings'])\n", "\n", "# 获取特征feature以及类别的label\n", "temp = extract_features(parent_dir,sub_dirs,max_file=100)" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "temp = np.array(temp)\n", "data = temp.transpose()" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# 获取特征\n", "X = np.vstack(data[:, 0])\n", "\n", "# 获取标签\n", "Y = np.array(data[:, 1])\n", "print('X的特征尺寸是：',X.shape)\n", "print('Y的特征尺寸是：',Y.shape)" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# 在Keras库中：to_categorical就是将类别向量转换为二进制（只有0和1）的矩阵类型表示\n", "Y = to_categorical(Y)" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "'''最终数据'''\n", "print(X.shape)\n", "print(Y.shape)" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "X_train, X_test, Y_train, Y_test = train_test_split(X, Y, random_state = 1, stratify=Y)\n", "print('训练集的大小',len(X_train))\n", "print('测试集的大小',len(X_test))" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "X_train = X_train.reshape(-1, 16, 8, 1)\n", "X_test = X_test.reshape(-1, 16, 8, 1)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 搭建CNN网络¶\n" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "model = Sequential()\n", "\n", "# 输入的大小\n", "input_dim = (16, 8, 1)\n", "\n", "model.add(Conv2D(64, (3, 3), padding = \"same\", activation = \"tanh\", input_shape = input_dim))# 卷积层\n", "model.add(MaxPool2D(pool_size=(2, 2)))# 最大池化\n", "model.add(Conv2D(128, (3, 3), padding = \"same\", activation = \"tanh\")) #卷积层\n", "model.add(MaxPool2D(pool_size=(2, 2))) # 最大池化层\n", "model.add(Dropout(0.1))\n", "model.add(Flatten()) # 展开\n", "model.add(Dense(1024, activation = \"tanh\"))\n", "model.add(Dense(20, activation = \"softmax\")) # 输出层：20个units输出20个类的概率\n", "\n", "# 编译模型，设置损失函数，优化方法以及评价标准\n", "model.compile(optimizer = 'adam', loss = 'categorical_crossentropy', metrics = ['accuracy'])" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "model.summary()" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# 训练模型\n", "model.fit(X_train, Y_train, epochs = 20, batch_size = 15, validation_data = (X_test, Y_test))" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "#预测\n", "\n", "def extract_features(test_dir, file_ext=\"*.wav\"):\n", " feature = []\n", " for fn in tqdm(glob.glob(os.path.join(test_dir, file_ext))[:]): # 遍历数据集的所有文件\n", " X, sample_rate = librosa.load(fn,res_type='kaiser_fast')\n", " mels = np.mean(librosa.feature.melspectrogram(y=X,sr=sample_rate).T,axis=0) # 计算梅尔频谱(mel spectrogram),并把它作为特征\n", " feature.extend([mels])\n", " return feature\n", " " ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "X_test = extract_features('./test_a/')" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "X_test = np.vstack(X_test)\n", "predictions = model.predict(X_test.reshape(-1, 16, 8, 1))" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "preds = np.argmax(predictions, axis = 1)\n", "preds = [label_dict_inv[x] for x in preds]\n", "\n", "path = glob.glob('./test_a/*.wav')\n", "result = pd.DataFrame({'name':path, 'label': preds})\n", "\n", "result['name'] = result['name'].apply(lambda x: x.split('/')[-1])\n", "result.to_csv('submit.csv',index=None)" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "!ls ./test_a/*.wav | wc -l" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "!wc -l submit.csv" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [] } ], "metadata": { "language_info": { "name": "plaintext" }, "orig_nbformat": 4 }, "nbformat": 4, "nbformat_minor": 2 }