International Research Journal of Engineering and Technology

2.1 Audio classification and segmentation steps 
Hybrid classification scheme is proposed that allows you to 
classify an audio clip into simple statistics sorts. Before type a 
pre-classification step is done which analyzes each windowed 
body of the audio clip separately. Then the feature extraction 
step is carried out from which a normalized feature vector is 
acquired. After feature extraction the hybrid classifier 
approach is used. The first step classifies audio clips/frames 
into natural-eco and noise segments via the use of bagged 
SVM. As the silence frames are mostly found in audio signal 
so the frequency section is assessed into silence and pure-eco 
segments on the idea of rule-based totally classifier. 
2.2 Pre-classification 
Digital signal is superimposed (i.e., in combined shape) 
which means that a communication is held at any location or 
celebration where there's echoes and plenty of noise. This is 
likewise known as cocktail celebration impact. Separating the 
source or the desired segments in the independent thing 
analysis framework is known as blind source separation. 
Blind supply is usually a method used to separate the 
combined sign into independent sources (while the 
combination process isn't always recognized) [. Most blind 
supply separation techniques use better order statistics. For 
higher order facts these algorithms require iterative 
calculations. This does no longer want better order data and 
iterative calculations. The temporal shape of alerts is analyzed 
and the separation is done on this foundation. 
The mixed sign is first of all converted to the time-frequency 
domain, additionally known as spectrogram of sign, via 
making use of Fourier remodel at short-time periods. 
Hamming window is used. In order to avoid blending of 
spectrograms each spectrogram is handled one by one. 
Correlation is performed on a majority of these short 
durations. A statement is simply a projection of supply alerts 
in positive direction. Reconstruction step is achieved on each 
separated sign’s spectrogram. All the decomposed frequency 
components are then combined. At the end permutation step 
is done for finding the relation between the separated 
indicators. The choice is made with the aid of using classifier.