python save不同名称_语音识别训练——遍历文件夹实现音频数据扩增（Data Augmentation) Python 代码...

· 如何实现数据增强（Data Augmentation）?

语音识别，需要大量的数据样本，试验中收集的样本个数有限，可以采用数据增强的方式扩增数据，而不改变数据中原有的信息。

音频数据常见的数据增强方式有：加噪，Shifting, Stretching

Add Noise Wave Plot

Shifting and Stretching Wave Plot

加噪的Python 代码如下：

#coding=gbk

import os

import librosa

import numpy as np


def Add_noise(data,p):

    wn = np.random.normal(0,1,len(data))

    data_noise = p*wn+data

    return data_noise
    


path = 'E:/xxx/xxx/' 
files = os.listdir(path)
files = [path + f for f in files if f.endswith('.wav')]


for i in range(len(files)): 

    
    # 加载需要进行加噪声扩增的音频
    
    FileName = files[i]

    print("add noise File Name is ", FileName)
    
    
    y, sr = librosa.load(files[i], sr=None)  
    
    t = librosa.get_duration(y=y, sr=sr)
    
    print(t)
    
    

    # 设置加噪声扩增后的wav文件保存名称及保存路径
    
    save_name = FileName[105:-4]+'-noise'+'.wav'
    
    print(save_name)
    
    save_path = "E:xxxxxx" 
    
    path_noise = save_path+save_name
    
    print(path_noise)
    
    
    
    # 加入噪声，进行信号扩增
    
    p = 0.02 #加入0.02*高斯白噪声 
    
    data_noise = Add_noise(y, p)

   

    # Saving the audio 
    
    librosa.output.write_wav(path_noise, data_noise, sr)
 

print('run over！')

Shifting 和 Stretching的代码如下：

#coding=gbk

import os
import librosa
import numpy as np
import random
import matplotlib.pyplot as plt
import librosa.display
%matplotlib inline



def Shift_Wave(y,sr):
    
    n_steps = random.randint(-10, 10)
    
    y_ps = librosa.effects.pitch_shift(y, sr, n_steps) #核心参数：n_steps
    
    plt.subplot(511)

    librosa.display.waveplot(y_ps,sr=sr)

    plt.title('Pitch Shift transformed waveform')
    
    return y_ps


def Stretch_Wave(y):  

    rate = random.uniform(0.8, 1.2)
    
    y_ts = librosa.effects.time_stretch(y, rate) #核心参数：rate

    plt.subplot(513)
    
    librosa.display.waveplot(y_ts,sr=sr)

    plt.title('Time Stretch transformed waveform')
    
    return y_ts


    
path = 'E:/xxx/xxx/' 
files = os.listdir(path)
files = [path + f for f in files if f.endswith('.wav')]

    
for i in range(len(files)): 

    
    # 加载需要进行数据扩增的音频
    
    FileName = files[i]

    print("Shift Wave File Name is ", FileName)
    
    y, sr = librosa.load(files[i], sr=None)  
    
    plt.subplot(515)
    
    librosa.display.waveplot(y,sr=sr)
  

    # 设置Shift扩增后的wav文件保存名称及保存路径
    
    save_name = FileName[105:-4]+'-shift'+'.wav'
    
    print(save_name)
    
    save_path_shift = "E:xxxxxx" 
    
    path_shift = save_path_shift+save_name
    
    print(path_shift)
      
    
    # 进行shift信号扩增
    
    data_shift = Shift_Wave(y,sr)
    
    
    
    # 设置stretch扩增后的wav文件保存名称及保存路径
    
    save_name = FileName[105:-4]+'-stretch'+'.wav'
    
    print(save_name)
    
    save_path_stretch = "E:xxxxxx" 
    
    path_stretch = save_path_stretch+save_name
    
    print(path_stretch)
    
    # 进行stretch信号扩增
    
    data_stretch = Stretch_Wave(y)

   

    # Saving the audio 
    
    librosa.output.write_wav(path_shift, data_shift, sr)
    
    librosa.output.write_wav(path_stretch, data_stretch, sr)
    

print('run over！')