# LPC and LSF

Started by May 24, 2005
```Dear Firndz

I have one querry regarding the LPC and LSF parameters

Why we compute the LSF paramenter when we already had computed the LPC

coefficeint. As far as I know all the relevent info is embeded in the
LPC coefficent then why we compute the LSF parameters ?

Only on thing I can annalyis is that Numbers of LPC coefficent can
be represented by the less neumber of LSF parameters..
example 10 LPC coefficent can be represnted with 5 LSF coefficent,

So we compute the LSF parameter to avoid the extra mathematical
computation ??

One more thing I have seen in most of the case we go for the FIR
filters,
with cutt off frequency with 1khz .... how we get this magic number ??

why we cant go ahead with the IIR Filters ?

what is the diffrence between the FIR filters and the IIR Filters ?

these queries.

Regards
don

```
```Don wrote:
> Dear Firndz
>
>  I have one querry regarding the LPC and LSF parameters
>  Why we compute the LSF paramenter when we already had computed the LPC
>  coefficeint. As far as I know all the relevent info is embeded in the
>  LPC coefficent then why we compute the LSF parameters ?

Is this a homework problem?

The LSF parameters are usually easier to quantize and have better error
characteristics.  (If there's a bit error in the LSFs, it's easier to
detect and if you don't it doesn't hurt quality as much.  Sort of.)

>
>  Only on thing I can annalyis is that Numbers of LPC coefficent can
>  be represented by the less neumber of LSF parameters..
>  example 10 LPC coefficent can be represnted with 5 LSF coefficent,
>
>  So we compute the LSF parameter to avoid the extra mathematical
> computation ??
>
>  One more thing I have seen in most of the case we go for the FIR
> filters,
>  with cutt off frequency with 1khz .... how we get this magic number ??

Probably because most of the information that we want to extract are in
this band.  But I think the higher frequencies are still needed for
intelligibility.

>
>  why we cant go ahead with the IIR Filters ?
>

Probably could.  You don't say what filter you're talking about here.

Lots of hand-waving here, but I hope it helps a little.  Plus, I don't
remember all the details anymore.

Ray
```
```The LSF are parameters that are equivalent to the LPC.  They form an
alternative representation for the same information represented by the
LPC.  The same number of coefficients is used in both representations.
But, LSF are more desirable for quantization and interpolation.

FIR = Finite Impulse Response
IIR = Infinite Impulse Response
That's perhaps the most obvious difference between the two.

Don wrote:
> Dear Firndz
>
>  I have one querry regarding the LPC and LSF parameters
>
>  Why we compute the LSF paramenter when we already had computed the LPC
>
>  coefficeint. As far as I know all the relevent info is embeded in the
>  LPC coefficent then why we compute the LSF parameters ?
>
>  Only on thing I can annalyis is that Numbers of LPC coefficent can
>  be represented by the less neumber of LSF parameters..
>  example 10 LPC coefficent can be represnted with 5 LSF coefficent,
>
>  So we compute the LSF parameter to avoid the extra mathematical
> computation ??
>
>  One more thing I have seen in most of the case we go for the FIR
> filters,
>  with cutt off frequency with 1khz .... how we get this magic number ??
>
>  why we cant go ahead with the IIR Filters ?
>
>  what is the diffrence between the FIR filters and the IIR Filters ?
>
>  Please help me as I am not a DSP GUY but definetly will like to be
>  these queries.
>
>  Regards
>  don
>
```
```Hi there
"The LSF parameters are usually easier to quantize and have better error
characteristics. (If there's a bit error in the LSFs, it's easier to detect
and if you don't it doesn't hurt quality as much. Sort of.)"

I want to know is it true " LSF coeffs are more roboust against the
channel noise"? can you compare LPC , PARCOR and LSF coeffs and say which
one is more roboust against the noise of channel? can you lead me to a
paper to get answer for this question?