基础语音识别-食物语音识别baseline(CNN)_Python

MFCC

梅尔倒谱系数（Mel-scaleFrequency Cepstral Coefficients，简称MFCC）。

MFCC通常有以下之过程:

将一段语音信号分解为多个讯框。
将语音信号预强化，通过一个高通滤波器。
进行傅立叶变换，将信号变换至频域。
将每个讯框获得的频谱通过梅尔滤波器(三角重叠窗口)，得到梅尔刻度。
在每个梅尔刻度上提取对数能量。
对上面获得的结果进行离散傅里叶反变换，变换到倒频谱域。
MFCC就是这个倒频谱图的幅度(amplitudes)。一般使用12个系数，与讯框能量叠加得13维的系数。

基础语音识别-食物语音识别baseline(CNN)

数据集

数据集来自Eating Sound Collection，数据集中包含20种不同食物的咀嚼声音，赛题任务是给这些声音数据建模，准确分类。

类别包括: aloe, ice-cream, ribs, chocolate, cabbage, candied_fruits, soup, jelly, grapes, pizza, gummies, salmon, wings, burger, pickles, carrots, fries, chips, noodles, drinks

训练集的大小: 750

测试集的大小: 250

1 下载和解压数据集

!wget http://tianchi-competition.oss-cn-hangzhou.aliyuncs.com/531887/train_sample.zip
!unzip -qq train_sample.zip
!
m train_sample.zip
!wget http://tianchi-competition.oss-cn-hangzhou.aliyuncs.com/531887/test_a.zip
!unzip -qq test_a.zip
!
m test_a.zip

2 加载库函数

# 基本库
import pandas as pd
import numpy as np
from sklearn.model_selection import train_test_split  #划分数据集
from sklearn.metrics import classification_report   #用于显示主要分类指标的文本报告
from sklearn.model_selection import GridSearchCV #自动调参
from sklearn.preprocessing import MinMaxScaler #归一化

加载深度学习框架

# 搭建分类模型所需要的库
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Conv2D, Flatten, Dense, MaxPool2D, Dropout
from tensorflow.keras.utils import to_categorical 
from sklearn.ensemble import RandomForestClassifier
from sklearn.svm import SVC #支持向量分类
!pip install librosa --user #加载音频处理库
# 其他库
import os
import librosa #音频处理库
import librosa.display
import glob

3 特征提取以及数据集的建立

建立类别标签字典

feature = []
label = []
# 建立类别标签，不同类别对应不同的数字。
label_dict = {"aloe": 0, "burger": 1, "cabbage": 2,"candied_fruits":3, "carrots": 4, "chips":5,
                  "chocolate": 6, "drinks": 7, "fries": 8, "grapes": 9, "gummies": 10, "ice-cream":11,
                  "jelly": 12, "noodles": 13, "pickles": 14, "pizza": 15, "ribs": 16, "salmon":17,
                  "soup": 18, "wings": 19}
label_dict_inv = {v:k for k,v in label_dict.items()}

提取梅尔频谱特征

from tqdm import tqdm
def extract_features(parent_dir, sub_dirs, max_file=10, file_ext="*.wav"):
    c = 0
    label, feature = [], []
    for sub_dir in sub_dirs:
        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
           # segment_log_specgrams, segment_labels = [], []
            #sound_clip,sr = librosa.load(fn)
            #print(fn)
            label_name = fn.split("/")[-2]
            label.extend([label_dict[label_name]])
            X, sample_rate = librosa.load(fn,res_type="kaiser_fast")
            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
            feature.extend([mels])
    return [feature, label]

# 自己更改目录
parent_dir = "./train_sample/"
save_dir = "./"
folds = sub_dirs = np.array(["aloe","burger","cabbage","candied_fruits",
                             "carrots","chips","chocolate","drinks","fries",
                            "grapes","gummies","ice-cream","jelly","noodles","pickles",
                            "pizza","ribs","salmon","soup","wings"])
# 获取特征feature以及类别的label
temp = extract_features(parent_dir,sub_dirs,max_file=100)
temp = np.array(temp)
data = temp.transpose()

获取特征和标签

# 获取特征
X = np.vstack(data[:, 0])
# 获取标签
Y = np.array(data[:, 1])
print("X的特征尺寸是：",X.shape)
print("Y的特征尺寸是：",Y.shape)

X的特征尺寸是： (1000, 128)

Y的特征尺寸是： (1000,)

独热编码

# 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
Y = to_categorical(Y)
print(X.shape)
print(Y.shape)

(1000, 128)

(1000, 20)

把数据集划分为训练集和测试集

X_train, X_test, Y_train, Y_test = train_test_split(X, Y, random_state = 1, stratify=Y)
print("训练集的大小",len(X_train))
print("测试集的大小",len(X_test))

训练集的大小 750

测试集的大小 250

X_train = X_train.reshape(-1, 16, 8, 1)
X_test = X_test.reshape(-1, 16, 8, 1)

4 建立模型

搭建CNN网络

model = Sequential()
# 输入的大小
input_dim = (16, 8, 1)
model.add(Conv2D(64, (3, 3), padding = "same", activation = "tanh", input_shape = input_dim))# 卷积层
model.add(MaxPool2D(pool_size=(2, 2)))# 最大池化
model.add(Conv2D(128, (3, 3), padding = "same", activation = "tanh")) #卷积层
model.add(MaxPool2D(pool_size=(2, 2))) # 最大池化层
model.add(Dropout(0.1))
model.add(Flatten()) # 展开
model.add(Dense(1024, activation = "tanh"))
model.add(Dense(20, activation = "softmax")) # 输出层：20个units输出20个类的概率
# 编译模型，设置损失函数，优化方法以及评价标准
model.compile(optimizer = "adam", loss = "categorical_crossentropy", metrics = ["accuracy"])
model.summary()

基础语音识别-食物语音识别baseline(CNN)

训练模型

# 训练模型
model.fit(X_train, Y_train, epochs = 100, batch_size = 15, validation_data = (X_test, Y_test))

5 预测测试集

def extract_features(test_dir, file_ext="*.wav"):
    feature = []
    for fn in tqdm(glob.glob(os.path.join(test_dir, file_ext))[:]): # 遍历数据集的所有文件
        X, sample_rate = librosa.load(fn,res_type="kaiser_fast")
        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
        feature.extend([mels])
    return feature
X_test = extract_features("./test_a/")
X_test = np.vstack(X_test)
predictions = model.predict(X_test.reshape(-1, 16, 8, 1))
preds = np.argmax(predictions, axis = 1)
preds = [label_dict_inv[x] for x in preds]
path = glob.glob("./test_a/*.wav")
result = pd.DataFrame({"name":path, "label": preds})
result["name"] = result["name"].apply(lambda x: x.split("/")[-1])
result.to_csv("submit.csv",index=None)
!ls ./test_a/*.wav | wc -l
!wc -l submit.csv