Forums

Question regarding G.726 speech codec

Started by Padmalatha Y July 4, 2003
Hi All,

I have few doubts regarding Homing that is done in G.726. The ITU test
vectors have few homing sequence for which they have few initialization files.
The document says homing is done to initialize the application. Now my doubt is
in real time say for ex when we speak if it is passing through my g.726 at
starting it will be in reset or in homing,if it is homing can you plz explain
the concept of it. That is in the homing it is not passing through my init ,so
in real time how does it works every time when I speak before that some samples
or runned or the previous thing will be stored in that state variables.
Can you plz tell me about it.
Homing is only for testing or it is one of the necessary thing.



Hi all,
Any one of you know information regarding ER AAC Scalable. I have ISO/IEC Specs
with me. I want more information regarding Decoder. Please suggest me any papers
on "BIT STREAM PARSING" part in Decoder implimentation.

Regards,
Ramya.




hi All...:)

anyone know about how to make a codebook in Vector Quantization used in CELP
coding ? how to implement Linde, Buzo, Gray (LBG) Algorithm ?

thx b4.......

regards,
arida



First, I assume you are talking about the fix codebook in CELP
coding(another one is adaptive codebook which is just a pitch
predictor..). There are all kinds of fixed-codebook used in CELP, i.e.:
Algebraic codebook: it is a highly structured codebook with rules on pulse
position, # of pulses each frame, and signs. As used in G.729/G.723 low
rate. There are also other structured codebooks similar to this, the
purpose is to save complexity for fixed-codebook search.
Random codebook: just generate a bunch of random vectors with unit power.
Other trained codebooks: in order to generate such codebook,you can use
closed-loop training or open-loop training. The open-loop training is like
any other common VQ codebook training, you first generate a trainig
databsae consisting of the residual vectors in the CELP coder, then you
can use all kinds of general VQ training method, i.e. GLA method. The best
reference book on this is by Allen Gersho's "Vector Quantization". The
closed-loop codebook trainig means the training vectors are your frames of
real speech. You embeded the whole trainig process inside your CELP coding
procedure. The complexity is pretty high in this closed-loop training, but
supposely you get better performed codebooks.

Thanks,

-Chunyan

>
>
> hi All...:)
>
> anyone know  about how to make a codebook in Vector Quantization
> used in CELP coding ? how to implement Linde, Buzo, Gray (LBG) Algorithm
> ?
>
> thx b4.......
>
> regards,
> arida > _____________________________________
> Note: If you do a simple "reply" with your email client, only
> the author of this message will receive your answer.  You need to
> do a "reply all" if you want your answer to be distributed to
> the entire group.
>
> _____________________________________
> About this discussion group:
>
> To Join: 
>
> To Post: 
>
> To Leave:
>
> Archives: http://www.yahoogroups.com/group/speechcoding
>
> Other DSP-Related Groups: http://www.dsprelated.com





Hi,

As a briefing has already been given...i'll give the
book that i reffered..(others were all papers by
indviduals)..and found it a bit usefull..

Digital Speech - A.M.Kondoz

Bye,
Prasadh ramachandran
--- wrote:
> First, I assume you are talking about the fix
> codebook in CELP
> coding(another one is adaptive codebook which is
> just a pitch
> predictor..). There are all kinds of fixed-codebook
> used in CELP, i.e.:
> Algebraic codebook: it is a highly structured
> codebook with rules on pulse
> position, # of pulses each frame, and signs. As used
> in G.729/G.723 low
> rate. There are also other structured codebooks
> similar to this, the
> purpose is to save complexity for fixed-codebook
> search.
> Random codebook: just generate a bunch of random
> vectors with unit power.
> Other trained codebooks: in order to generate such
> codebook,you can use
> closed-loop training or open-loop training. The
> open-loop training is like
> any other common VQ codebook training, you first
> generate a trainig
> databsae consisting of the residual vectors in the
> CELP coder, then you
> can use all kinds of general VQ training method,
> i.e. GLA method. The best
> reference book on this is by Allen Gersho's "Vector
> Quantization". The
> closed-loop codebook trainig means the training
> vectors are your frames of
> real speech. You embeded the whole trainig process
> inside your CELP coding
> procedure. The complexity is pretty high in this
> closed-loop training, but
> supposely you get better performed codebooks.
>
> Thanks,
>
> -Chunyan
>
> >
> >
> >
> >
> >
> > hi All...:)
> >
> > anyone know  about how to make a codebook in
> Vector Quantization
> > used in CELP coding ? how to implement Linde,
> Buzo, Gray (LBG) Algorithm
> > ?
> >
> > thx b4.......
> >
> > regards,
> > arida
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> > _____________________________________
> > Note: If you do a simple "reply" with
> your email client, only
> > the author of this message will receive your
> answer.  You need to
> > do a "reply all" if you want your answer
> to be distributed to
> > the entire group.
> >
> > _____________________________________
> > About this discussion group:
> >
> > To Join: 
>
> >
> > To Post: 
> >
> > To Leave:
> >
> > Archives:
> http://www.yahoogroups.com/group/speechcoding
> >
> > Other DSP-Related Groups:
> http://www.dsprelated.com
> >
> >
> >
> >
>


__________________________________