音频.wav处理

Halcom

需要执行HMM的朋友，请自行安装：python_speech_features，请参考http://python-speech-features.readthedocs.org/en/latest.
使用环境：Win7-32bit-Anaconda2-4.3.1-Windows-x86.exe

1. import numpy as np
   
2. from scipy.io import wavfile
   
3. import matplotlib.pyplot as plt
   
4. 
   
5. # Read the input file
   
6. sampling_freq, audio = wavfile.read('input_freq.wav')
   
7. 
   
8. # Normalize the values
   
9. audio = audio / (2.**15)
   
10. 
    
11. # Extract length
    
12. len_audio = len(audio)
    
13. 
    
14. # Apply Fourier transform
    
15. transformed_signal = np.fft.fft(audio)
    
16. half_length = np.ceil((len_audio + 1) / 2.0)
    
17. transformed_signal = abs(transformed_signal[0:half_length])
    
18. transformed_signal /= float(len_audio)
    
19. transformed_signal **= 2
    
20. 
    
21. # Extract length of transformed signal
    
22. len_ts = len(transformed_signal)
    
23. 
    
24. # Take care of even/odd cases
    
25. if len_audio % 2:
    
26.     transformed_signal[1:len_ts] *= 2
    
27. else:
    
28.     transformed_signal[1:len_ts-1] *= 2
    
29. 
    
30. # Extract power in dB
    
31. power = 10 * np.log10(transformed_signal)
    
32. 
    
33. # Build the time axis
    
34. x_values = np.arange(0, half_length, 1) * (sampling_freq / len_audio) / 1000.0
    
35. 
    
36. # Plot the figure
    
37. plt.figure()
    
38. plt.plot(x_values, power, color='black')
    
39. plt.xlabel('Freq (in kHz)')
    
40. plt.ylabel('Power (in dB)')
    
41. plt.show()

复制代码

参考：https://github.com/PacktPublishing/Python-Machine-Learning-Cookbook