Re: abt VAD algorithm 1 of AMR codec

Started by October 19, 2006
hi sameer,
do you still have the c-code for the vad you mentioned below. i'm working on it now and it would be quite helpful if you could help me out with it.
looking forward to it. if anyother member has it.... plz forward it to me ..

thank you,

rakesh
>
>
>
> Hi all!
>
>i have a query regarding the VAD algorithm 1 used in GSM AMR speech codec.one of
>the steps in this algo is the use of a filter bank. this bank consists of 5th
>and 3rd order filter blocks.
>
>for the 3rd order filter block the equations are as follows :
>
>x_lp(n) = 0.5 * (x(n) + A3(x(n-1))) eq (3.2a)
>x_hp(n) = 0.5 * (x(n) - A3(x(n-1))) eq (3.2b)
>
>where
>x(n) is the input signal
>x_lp(n) = low pass component
>x_hp(n) = high pass component
>
>A3() is said to be a first order direct form all-pass filter, whose transfer
>function is given by :
>
>A3(z) = [C3+z^-1]/[1+Cz^-1]
>where C3 = filter coefft
>
>now, the term A3 in eq 3.2a and 3.2b are confusing since all the other terms are
>in time domain. in fact, going by their C code, i found that the term is
>C3*x(n-1) and not A3(x(n-1)).
>
>even for 5th order filter, same results are obtained.(A1 to be replaced by C1
>etc).
>is this observation correct?if it is then there is a printing mistake in the
>standard which is generally unlikely.
>if anyone has worked on this earlier please let me know the solution.
>
>Thank you
>Sameer Kibey
>
>
>
>
>
>