Forums

50th order LPC

Started by Bhanu Prakash May 7, 2002
Hi Hari,

What ever you have said may be applicable for processing music signals(may be
MP3).But G.728 is a speech codec whose inputs are sampled at 8khz .As a matter
of fact,all the basic speech codecs expect input signals which are sampled at
8KHz. So your answer will not be applicable in this context. Also for speech
coding,Toll quality would suffice.

I guess that 50th order LPC is to model not only the formants but also the fine
structure of the speech spectrum. But i am not sure....????

Any comments...get back ..

Best regards,
Bhanu Prakash.

-------Original Message-------

From: Harikrishna Natarajan
Date: Monday, May 06, 2002 11:41:00 PM
To:
Subject: Re: [speechcoding] All pole LPC filter again

Hi

Great to have a good follow-up on an interesting topic. It would be great if
more who have good exposure to this topic, join in.

>All the reasonings given for 10th order LPC is plausible. But i
>encountered
>50th order filter
>in G.728(LD-CELP). what could be the probable reason for having such a
>high
>order
>(computationally intensive) LPC in this codec.
>
>Note: The codec is a backward adaptive and uses 20 sample frame(2.5msec).

The choice of the order of the filter basically depends on the sampling
frequency (to bo more precise, half sampling frequency). Its usually taken
as follows:

A complex pole for every KHZ in half the sampling frequency and 2-4 extra
poles. So for CD quality sounds, with 22.1 KHZ we need about 23 cplx poles
(46 order) + 4 extra poles = about 50.

Any comments?

-Hari



Hi all!

i haven't worked in G.728, but my understanding is
that the use of 50th order prediction filter in
LD-CELP makes it possible to incorporate the long term
prediction (pitch info) along with the short term
prediction(formant info) with the same
predictor...(you see, the predictor is taking into
consideration past 50 samples instead of 10 this time
for the prediction)

in 10th order LPC analysis, the long term prediction
does not come into picture...it is a separate block.

please do not ask me for more clarifications this
time! for more details, however one can refer the
topic on LD-CELP given in the book by Kondoz.

Sameer.

--- Bhanu Prakash <> wrote: >
Hi Hari,
>
> What ever you have said may be applicable for
> processing music signals(may be MP3).But G.728 is a
> speech codec whose inputs are sampled at 8khz .As a
> matter of fact,all the basic speech codecs expect
> input signals which are sampled at 8KHz. So your
> answer will not be applicable in this context. Also
> for speech coding,Toll quality would suffice.
>
> I guess that 50th order LPC is to model not only the
> formants but also the fine structure of the speech
> spectrum. But i am not sure....????
>
> Any comments...get back ..
>
> Best regards,
> Bhanu Prakash. >
>
> -------Original Message-------
>
> From: Harikrishna Natarajan
> Date: Monday, May 06, 2002 11:41:00 PM
> To:
> Subject: Re: [speechcoding] All pole LPC filter
> again
>
> Hi
>
> Great to have a good follow-up on an interesting
> topic. It would be great if
> more who have good exposure to this topic, join in.
>
> >All the reasonings given for 10th order LPC is
> plausible. But i
> >encountered
> >50th order filter
> >in G.728(LD-CELP). what could be the probable
> reason for having such a
> >high
> >order
> >(computationally intensive) LPC in this codec.
> >
> >Note: The codec is a backward adaptive and uses 20
> sample frame(2.5msec).
>
> The choice of the order of the filter basically
> depends on the sampling
> frequency (to bo more precise, half sampling
> frequency). Its usually taken
> as follows:
>
> A complex pole for every KHZ in half the sampling
> frequency and 2-4 extra
> poles. So for CD quality sounds, with 22.1 KHZ we
> need about 23 cplx poles
> (46 order) + 4 extra poles = about 50.
>
> Any comments?
>
> -Hari >
> ------------------------ Yahoo! Groups Sponsor
>
> _____________________________________
> 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 > ">http://docs.yahoo.com/info/terms/

__________________________________________________



Hi all,
It good to see 50th order LPC debate is still on.I
agree with wthat sameer has said.As far as i know the
earlier versions of LD_CELP used to have a 10th order
backward LPC analysis and a backward adaptive pitch
detector. The main problem with such a configuration
is its poor performance under channel noise as its
difficult to detect actual pitch in noisy
conditions.The inclusion of a LTP makes the system
very sensitive to channel errors.Therefore, the LTP
was ommited and instead a high oreder LPC driven STP
was included.The inclusion of a 50th order backward
STP has the following advantages:
High order backward LPC analysis is quite robust to
channel errors.
No side information is needed for higher orders.
The algorithm becomes less speech specific by
replacing the LTP with high order STP,hence it is
cabable of handling non-voice signals.
The disadvantages are:
A 50th oreder LPC requires a large number of
correlation compution , thus resulting in high
complexity.
regs
naveen
--- Sameer Kibey <> wrote:
> Hi all!
>
> i haven't worked in G.728, but my understanding is
> that the use of 50th order prediction filter in
> LD-CELP makes it possible to incorporate the long
> term
> prediction (pitch info) along with the short term
> prediction(formant info) with the same
> predictor...(you see, the predictor is taking into
> consideration past 50 samples instead of 10 this
> time
> for the prediction)
>
> in 10th order LPC analysis, the long term prediction
> does not come into picture...it is a separate block.
>
> please do not ask me for more clarifications this
> time! for more details, however one can refer the
> topic on LD-CELP given in the book by Kondoz.
>
> Sameer.
>
> --- Bhanu Prakash <> wrote:
> >
> Hi Hari,
> >
> > What ever you have said may be applicable for
> > processing music signals(may be MP3).But G.728 is
> a
> > speech codec whose inputs are sampled at 8khz .As
> a
> > matter of fact,all the basic speech codecs expect
> > input signals which are sampled at 8KHz. So your
> > answer will not be applicable in this context.
> Also
> > for speech coding,Toll quality would suffice.
> >
> > I guess that 50th order LPC is to model not only
> the
> > formants but also the fine structure of the speech
> > spectrum. But i am not sure....????
> >
> > Any comments...get back ..
> >
> > Best regards,
> > Bhanu Prakash.
> >
> >
> >
> >
> > -------Original Message-------
> >
> > From: Harikrishna Natarajan
> > Date: Monday, May 06, 2002 11:41:00 PM
> > To:
> > Subject: Re: [speechcoding] All pole LPC filter
> > again
> >
> > Hi
> >
> > Great to have a good follow-up on an interesting
> > topic. It would be great if
> > more who have good exposure to this topic, join
> in.
> >
> > >All the reasonings given for 10th order LPC is
> > plausible. But i
> > >encountered
> > >50th order filter
> > >in G.728(LD-CELP). what could be the probable
> > reason for having such a
> > >high
> > >order
> > >(computationally intensive) LPC in this codec.
> > >
> > >Note: The codec is a backward adaptive and uses
> 20
> > sample frame(2.5msec).
> >
> > The choice of the order of the filter basically
> > depends on the sampling
> > frequency (to bo more precise, half sampling
> > frequency). Its usually taken
> > as follows:
> >
> > A complex pole for every KHZ in half the sampling
> > frequency and 2-4 extra
> > poles. So for CD quality sounds, with 22.1 KHZ we
> > need about 23 cplx poles
> > (46 order) + 4 extra poles = about 50.
> >
> > Any comments?
> >
> > -Hari
> >
> >
> >
> >
> >
> >
> > ------------------------ Yahoo! Groups Sponsor
> >
> > _____________________________________
> > 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
> >
> >
> > ">http://docs.yahoo.com/info/terms/
> >
> >
>
> __________________________________________________ >


__________________________________________________