From: ajay_n69 on
Hi,


I am trying to implement a high frequency regeneration system where I
take a narrowband speech signal and reconstruct high frequecies to
obtain a wideband signal. I have done LP analysis on the NB speech and
have LPC's and the residue. I now want to pass the residue (after
upsampling)through a whitening filter to obtain a residue which has a
flat spectrum. I just cant figure out how to implement a whitening
filter. As far as i understand, an LP analysis filter should do the job
but i dont know how to design one. Plz help, as I have been breaking my
head on this one. I am pretty new to matlab. I could post my code if
its necessary.....


PLZ PLZ PLZ help...


AJ

From: Lars Hansen on
If you have a frame of speech stored in the variable x the output of the
whitening filter will be

[a,e]=lpc(x,10)
a=real(a);
e=sqrt(e);

y=filter(a,e,x)

------
> flat spectrum. I just cant figure out how to implement a whitening
> filter. As far as i understand, an LP analysis filter should do the job
> but i dont know how to design one. Plz help, as I have been breaking my
> head on this one. I am pretty new to matlab. I could post my code if
> its necessary.....


From: abariska on
ajay_n69(a)hotmail.com wrote:
> I am trying to implement a high frequency regeneration system where I
> take a narrowband speech signal and reconstruct high frequecies to
> obtain a wideband signal. I have done LP analysis on the NB speech and
> have LPC's and the residue. I now want to pass the residue (after
> upsampling)through a whitening filter to obtain a residue which has a
> flat spectrum.

If you chose the correct order of the predictor, and the signal is
indeed a white noise excited AR process, then the residue is already
white. The predictor is the whitening filter.

From: Carlos Moreno on
abariska(a)student.ethz.ch wrote:
> ajay_n69(a)hotmail.com wrote:
>
>>I am trying to implement a high frequency regeneration system where I
>>take a narrowband speech signal and reconstruct high frequecies to
>>obtain a wideband signal. I have done LP analysis on the NB speech and
>>have LPC's and the residue. I now want to pass the residue (after
>>upsampling)through a whitening filter to obtain a residue which has a
>>flat spectrum.
>
>
> If you chose the correct order of the predictor, and the signal is
> indeed a white noise excited AR process, then the residue is already
> white. The predictor is the whitening filter.

Nit pick -- the predictor is actually the filter that gets the
"non-white" portion; 1 - P is the whitening filter, which is
the filter that computes the residual.

Carlos
--