基于UrbanSound8K与CNN的音频分类

  语音方面的资料不如图像识别的多，所以特地写了一份博客（并不如何严谨），希望可以帮到大家。

  我们需要实现10种语音的分类：冷气机，汽车喇叭，儿童玩耍，狗吠声，钻孔，发动机空转，枪射击，手持式凿岩机，警笛，街头音乐

  每个录音长度约为4s，被放在10个fold文件中。

  我们采用keras（可以简单的认为keras是前端，tensorflow是后端，类似于tensorflow是个库，我们使用keras调用它的api）实现模型搭建，使用librosa（Librosa是一个用于音频、音乐分析、处理的python工具包）来处理语音。

• 导入这几个库即可

  import keras
  from keras.layers import Activation, Dense, Dropout, Conv2D, Flatten, MaxPooling2D
  from keras.models import Sequentia
  import librosa
  import librosa.display
  import numpy as np
  import pandas as pd
  import random
  

• 读取csv文件

  data = pd.read_csv('metadata/UrbanSound8K.csv')
  valid_data = data[['slice_file_name', 'fold' ,'classID', 'class']][ data['end']-data['start'] >= 3 ]
  valid_data['path'] = 'fold' + valid_data['fold'].astype('str') + '/' + valid_data['slice_file_name'].astype('str')
  

• 读入wav文件

  from tqdm import tnrange, tqdm_notebook
  
  D=[]
  
  for row in tqdm_notebook(valid_data.itertuples()): 
      print(row.path)
      print(row.classID)
      y1, sr1 = librosa.load("audio/" + row.path, duration=2.97)  
      ps = librosa.feature.melspectrogram(y=y1, sr=sr1)
      if ps.shape != (128, 128): 
              continue
      D.append( (ps, row.classID) )
  

• 划分训练集和测试集，前7000个为训练集，7000以后为数据集

  dataset = D
  random.shuffle(dataset)
  
  train = dataset[:7000]
  test = dataset[7000:]
  
  X_train, y_train = zip(*train)
  X_test, y_test = zip(*test)
  
  X_train = np.array([x.reshape( (128, 128, 1) ) for x in X_train])
  X_test = np.array([x.reshape( (128, 128, 1) ) for x in X_test])
  
  y_train = np.array(keras.utils.to_categorical(y_train, 10))
  y_test = np.array(keras.utils.to_categorical(y_test, 10))
  

• 搭建模型

  model = Sequential()
  input_shape=(128, 128, 1)
  
  model.add(Conv2D(24, (5, 5), strides=(1, 1), input_shape=input_shape))
  model.add(MaxPooling2D((4, 2), strides=(4, 2)))
  model.add(Activation('relu'))
  
  model.add(Conv2D(48, (5, 5), padding="valid"))
  model.add(MaxPooling2D((4, 2), strides=(4, 2)))
  model.add(Activation('relu'))
  
  model.add(Conv2D(48, (5, 5), padding="valid"))
  model.add(Activation('relu'))
  
  model.add(Flatten())
  model.add(Dropout(rate=0.5))
  
  model.add(Dense(64))
  model.add(Activation('relu'))
  model.add(Dropout(rate=0.5))
  
  model.add(Dense(10))
  model.add(Activation('softmax'))
  

• 填入数据

  model.compile(
  optimizer="Adam",
  loss="categorical_crossentropy",
  metrics=['accuracy'])
  
  model.fit(
      x=X_train,
      y=y_train,
      epochs=12,
      batch_size=128,
      validation_data= (X_test, y_test))
  
  score = model.evaluate(
          x=X_test, y=y_test)
  
  print('Test loss:', score[0])
  print('Test accuracy:', score[1])
  

内容来自：https://blog.csdn.net/c2c2c2aa/article/details/81543549