In [31]:

    

%matplotlib inline
import os
os.environ["CUDA_VISIBLE_DEVICES"] = "-1"
import audio_utilities
import matplotlib.pyplot as plt
from IPython.display import Audio
import librosa
from librosa import display
import numpy as np
import sklearn
audio_utilities.ensure_audio()


    

In [3]:

    

SR = 16000
src_clean, _ = librosa.load('audio/clean/sp01.wav', sr=SR, mono=True)
src_noise, _ = librosa.load('audio/airport_10dB/sp01_airport_sn10.wav', sr=SR, mono=True)
print(SR)


    

    




16000

In [23]:

    

plt.plot(src_noise)
plt.plot(src_clean)


    

    
Out[23]:





[<matplotlib.lines.Line2D at 0x1163af1d0>]






    

In [29]:

    

librosa.stft(src_clean)
clean = librosa.amplitude_to_db(np.abs(librosa.stft(src_clean[:3*SR])), ref=np.max)
noise = librosa.amplitude_to_db(np.abs(librosa.stft(src_noise[:3*SR])), ref=np.max)
plt.figure(figsize=(16, 8))
plt.subplot(4, 2, 1)
display.specshow(clean, y_axis='log', sr=SR)
plt.colorbar(format='%+2.0f dB')
plt.subplot(4, 2, 2)
display.specshow(noise, y_axis='log',  sr=SR)
plt.colorbar(format='%+2.0f dB')
Audio(src_clean, rate=SR)


    

    
Out[29]:





                
              






    

In [26]:

    

Audio(src_noise, rate=SR)


    

    
Out[26]:





                
              

In [27]:

    

mfcc_clean = librosa.feature.mfcc(src_clean, sr=SR, n_mfcc=20)
mfcc_noise = librosa.feature.mfcc(src_noise, sr=SR, n_mfcc=20)
plt.figure(figsize=(16, 8))
plt.subplot(4, 2, 1)
display.specshow(mfcc_clean, x_axis='time')
plt.colorbar()
plt.subplot(4, 2, 2)
display.specshow(mfcc_noise, x_axis='time')


    

    
Out[27]:





<matplotlib.axes._subplots.AxesSubplot at 0x11c335b70>






    

In [32]:

    

display.specshow(sklearn.preprocessing.scale(mfcc_clean, axis=1))


    

    
Out[32]:





<matplotlib.axes._subplots.AxesSubplot at 0x106214400>






    

In [ ]: