Why non'symmetric window in LP based speech coders (for instance CS-ACELP G729...)

Hello,

I'm studying LP based coders but couldn't find proper explanation why non
symmetrical window (half Hamming and quarter of cosine) is used. Is it
because of magnitude response or time domain shape ?

What is scientific background to use of this window ?
How could I evaluate performance of some other window sequences - any
tutorials (preferably Matlab) available to do such evaluation ?


Thanks in advance,

Robert Rozman

Robert Rozman wrote:
> 
> Hello,
> 
> I'm studying LP based coders but couldn't find proper explanation why non
> symmetrical window (half Hamming and quarter of cosine) is used. Is it
> because of magnitude response or time domain shape ?
> What is scientific background to use of this window ?

There are two reasons:

1. The non-symmetric window emphasises the impact of the most recent
part of speech.

2. Some non-symmetric windows like Barnwell window allows to compute ACF
more efficiently.  
> How could I evaluate performance of some other window sequences - any
> tutorials (preferably Matlab) available to do such evaluation ?

There is indeed quite slight difference in perception if compared to
Hamming window.

Vladimir Vassilevsky

DSP and Mixed Signal Design Consultant

http://www.abvolt.com

Hello,

thanks for response, please see below:

"Vladimir Vassilevsky" <vlv@abvolt.com> wrote in message
news:3F4DFB51.5C3EA522@abvolt.com...
>
>
> There are two reasons:
>
> 1. The non-symmetric window emphasises the impact of the most recent
> part of speech.

This obviously is caused by time domain window shape right ?

>
> 2. Some non-symmetric windows like Barnwell window allows to compute ACF
> more efficiently.

Any references on that detail? What is ACF (Auto Correlation or something
else) ?

>
> There is indeed quite slight difference in perception if compared to
> Hamming window.

Are there any reference implementations available for research purposes and
suitable for such evaluation ?


Thanks in advance,

Robert Rozman

Robert Rozman wrote:
> 
> Hello,
> 
> thanks for response, please see below:
> 
> >
> > 1. The non-symmetric window emphasises the impact of the most recent
> > part of speech.
> 
> This obviously is caused by time domain window shape right ?

Yes.
 
> >
> > 2. Some non-symmetric windows like Barnwell window allows to compute ACF
> > more efficiently.
> 
> Any references on that detail? What is ACF (Auto Correlation or something
> else) ?

T.P. Barnwell "Recursive windowing for generating autocorrelation
coefficients for LPC analysis" 
IEEE Trans. on ASSP, Oct. 1981.


Vladimir Vassilevsky

DSP and Mixed Signal Design Consultant

http://www.abvolt.com