Forums

Pre-emphasize and First Formant

Started by Eris Ristemena September 5, 1999
Hi there,

can anyone explain to me about the use of pre-emphasize in LPC coding ?

If we see in the DOD Standard, they use pre-emphasize before the speech is
analyzed with 1 - 0.9375z**1.
I checked using MATLAB that the frequency response of this filter is high
pass, it means that we'll lose the first formant of the speech being
analyzed. If i've not mistaken, the first formant freq. is very important to
characterized the speech, so why they do the pre-emphasize? isnt they gonna
lose the intelegency of the analysis speech?

thanks

Regards,
Eris



wrote:
original article:http://www.egroups.com/group/speechcoding/?start%
> Hi there,
>
> can anyone explain to me about the use of pre-emphasize in LPC coding
?
>
> If we see in the DOD Standard, they use pre-emphasize before the
speech is
> analyzed with 1 - 0.9375z**1.
> I checked using MATLAB that the frequency response of this filter is
high
> pass, it means that we'll lose the first formant of the speech being
> analyzed. If i've not mistaken, the first formant freq. is very
important to
> characterized the speech, so why they do the pre-emphasize? isnt they
gonna
> lose the intelegency of the analysis speech?

Rabiner and Schafer (1978) page 433 gives a frequency
domaininterpretation of the standard LPC mean squared error. To
summarise I quote:

"it follows that minimising E_n (the mean squared error) is
equivalent to
minimising the ratio of the energy spectrum of the speech segment to
the
magnitude squared of the frequency response of the linear system in
the model of speech production"

The main point being that it is a ratio that is mimimised, so you can
play about with the power spectrum and to a first order approximation
the LPC model will still fit it. tony robinson
speaking only for