In [10]:

    

from pysoundfile import SoundFile
import matplotlib.pyplot as plt
import matplotlib as mpl
import numpy as np
import scipy, pylab
import copy
from mpltools import style
style.use('ggplot')
mpl.rcParams['figure.figsize'] = (18, 7)

#short time windowed fourier
def stft(x, framesamp, hop):
    w = scipy.hamming(framesamp)
    X = np.array([np.fft.rfft(w*x[i:i+framesamp]) for i in range(0, len(x)-framesamp, hop)],dtype=np.float64)
    #Note: might need to convert to phase and amplitude
    return X

#read wav
data = SoundFile("tickle_kitchen.wav")
#flatten and subsample
data = np.ravel(data)[::1000]
#short time fourier with 100 freqs (200/ nyquist freq)
out = stft(data,200,1)

#transpose for viewing
out = out.T

#freq band pass power summation - this need tuning as ther
outD = np.sum(out[0:20,:],axis=0)
#threshold 1.5 an arbritrary constant
outRaw = copy.deepcopy(outD) #deepcopy is housekeeping
outD[outD<2]=0
outD[outD>=2]=1
#plot - maybe a little phase lag in spectrogram (the middle plot) due to windowing
f, (ax1, ax2,ax3,ax4) = plt.subplots(4, 1, sharex=True)
ax1.plot(data) #first plot raw signal
ax2.imshow(out,aspect='auto') #spectrogram: frequency spectrum at a specific time
ax3.plot(outRaw) #band pass signal.... 
ax4.plot(outD) #just a simple threshold - clearly some analysis in the time domain would improve this
#misses the first noise but gets the end two - you get the picture though :)


    

    
Out[10]:





[<matplotlib.lines.Line2D at 0xd4fc6d0>]






    

In [5]:

    

#markov laughing

#markov p martix all rows sum to 1
decision = np.array([[0.2,0.3,0.5],
                     [0.3,0.4,0.3],
                     [0,8,0.1,0.1]])
#laugh states
event = ["hee","haa","hoo"]

#number of laughs
for i in range(10):
    lastMove = 0
    #length of laugh
    for j in range(3):
        #make a weighed random choice
        lastMove = np.random.choice(3,1,decision[lastMove,])
        print event[lastMove],
    print "\n"


    

    




hoo hoo hee 

hoo hee hoo 

hee haa hee 

haa haa hee 

hee hee haa 

haa haa hoo 

haa hee hee 

haa hoo haa 

hoo hoo haa 

hoo hoo hoo 

In [ ]: