

In [1]:

    
import numpy as np

import librosa

import scipy.io.wavfile

import senmf

import itertools

%pylab inline









    



Populating the interactive namespace from numpy and matplotlib






    



/usr/home/gsmith/work/senmf/lib/python2.7/site-packages/librosa/core.py:26: UserWarning: Could not import scikits.samplerate. Falling back to scipy.signal
  'Falling back to scipy.signal')

Compute a Spectrogram



In [2]:

    
sr, x = scipy.io.wavfile.read('data/bird-songs.wav')
stft = librosa.stft(x, n_fft=1024, hop_length=200).T

X = np.abs(stft)[:,:250]
figure(figsize=(20,5))
imshow(np.log(X.T))









    Out[2]:





<matplotlib.image.AxesImage at 0x828b4e510>

Decompose the Spectrogram



In [3]:

    
nmf = senmf.SENMF(n_bases=5, window_width=15, X=X)
nmf.rand_D()
nmf.rand_A()









    Out[3]:





array([[ 2.69193003,  2.25030021,  2.20126289, ...,  2.81231266,
         2.56364036,  2.09188649],
       [ 2.08217016,  2.58985266,  2.06753902, ...,  2.51650564,
         2.89342424,  2.05559205],
       [ 2.54388315,  2.03368312,  2.10184768, ...,  2.70425867,
         2.46019801,  2.19320678],
       [ 2.61670271,  2.39905904,  2.89251935, ...,  2.76419624,
         2.7481696 ,  2.14251139],
       [ 2.42815874,  2.87556784,  2.38494591, ...,  2.47523851,
         2.73166808,  2.67882737]])



In [5]:

    
%time nmf.fit(5)









    



CPU times: user 8.52 s, sys: 2.17 s, total: 10.7 s
Wall time: 10.7 s



In [6]:

    
plot(nmf.A.T)









    Out[6]:





[<matplotlib.lines.Line2D at 0x82c78e7d0>,
 <matplotlib.lines.Line2D at 0x82c78ea50>,
 <matplotlib.lines.Line2D at 0x82c78ec90>,
 <matplotlib.lines.Line2D at 0x82c78ee50>,
 <matplotlib.lines.Line2D at 0x82c79a050>]



In [7]:

    
(x,y,z) = nmf.D.shape
imshow(np.log(nmf.D.reshape(x*y, z)))









    Out[7]:





<matplotlib.image.AxesImage at 0x82c063310>

Inspect Decomposition



In [8]:

    
for i in range(nmf.n_bases):
    figure()
    imshow(nmf.reconstruct_basis(i))

Show the estimate



In [9]:

    
X_bar = nmf.reconstruct()
imshow(np.log(X_bar))









    Out[9]:





<matplotlib.image.AxesImage at 0x82c4a6c50>



In [10]:

    
imshow(np.log(X-X_bar))









    



/usr/home/gsmith/work/senmf/lib/python2.7/site-packages/IPython/kernel/__main__.py:1: RuntimeWarning: invalid value encountered in log
  if __name__ == '__main__':






    Out[10]:





<matplotlib.image.AxesImage at 0x82c55a950>