envelope_follower.m

function [silent_indices] = envelope_follower(x, ba)

%amplitude envelope tracker for data x
%ba - frame preceding or following silent frame
%1 - following(after), 0-preceding(before)

%Envelope Detection based on Hilbert Transform 

analy=hilbert(x);
env=abs(analy);
N = length(x);

%applying moving average filter to further smooth envelope

M = 10;
b = 1/M * ones(1,M);
a = 1;
env = filter(b,a,env);

%find signal indices with very low value of amplitude envelope
n = find(abs(env) < 0.5*max(abs(env)));
silent_indices = [];

if(~isempty(n))

    %there needs to be a consecutive array of silent indices. 
    %We will ignore single values that dip in amplitude, because they 
    %are most likely caused by computational errors. Consecutively 10
    %or more values need to be below the threshold for region to be silent.

    %the following lines find the longest sequence of consecutive numbers
    %in array n - very smart solution from Stack Overflow

    temp = [0 cumsum(diff(n)~=1)];
    elems = n(temp==mode(temp));

    %silent regions can only be at the beginning or end of frame, not in the
    %middle. The number of indices belonging to silent frames has to be
    %greater than 10

    if(numel(elems) >= 10)
        if(ba == 1)
            silent_indices = 1:max(elems);
        else
            silent_indices = min(elems):N;
        end
    end

end

end