International Research Journal of Engineering and Technology

2.4
 
Segmentation 
2.4.1
 
pyAudioAnalysis: 
[1] Silence removal. A semi-supervised silence removal 
functionality is also provided in the library. Their respective 
function takes an uninterrupted audio recording as input and 
returns segment endpoints that correspond to individual 
audio events, removing “silent” areas of the recording. This is 
achieved through a semi-supervised approach which 
performs the following steps: 
• The short
-term features of the whole recording are 
extracted 
• An
SVM model is trained to distinguish between high-
energy and low-energy short-term frames. In particular, 10% 
of the highest energy frames along with the10% of the lowest 
are used to train the SVM model 
• The SVM classifier is applied (with a probabilistic
output) 
on the whole recording, resulting in a sequence of 
probabilities that correspond to a level of confidence that 
their respective short-term frames belong to an audio event 
(and do not belong to a silent segment). 
• A dynamic thresholding is used to
detect the active 
segments. 
2.4.2
 
Pseudo code for segmentation 
3
[Fs,x]= 
aIO.readAudioFile("1904060514208888592030.wav") 
4
segments = aS.silenceRemoval(x, Fs, 0.025, 0.025,
smoothWindow = 0.85, plot = True) 
5
#type(segments) 
6
#print(Fs) Fs: the sampling rate of the generated WAV 
files 
7
#adding label 
8
n=1 
9
for i in segments:
10
#indexing of data to avoid using csv files and manual 
editing 
11
i.insert(0,n) 
12
#insert appends at the beginning
13
n=n+1 
14
print("The segmented wav files written are:") 
15
import scipy.io.wavfile as wavfile 
16
[Fs, 
x] 
= 
aIO.readAudioFile("1904060514208888592030.wav") 
17
for j in segments: 
18
T1 = float(j[1])#timestamp 1
19
T2 = float(j[2])#timestamp 2
20
label = ("1904060514208888592030.wav", j[0],
T1, T2) 
21
xtemp=x[int(round(T1*Fs)):int(round(T2*Fs))]
22
#time x (sapmle/time)
23
print (T1, T2, label, xtemp.shape) 
24
wavfile.write(label, Fs, xtemp)
25
print("the labeled limits are:\n",segments ) 

---