Newsgroups: comp.speech
Path: cantaloupe.srv.cs.cmu.edu!das-news.harvard.edu!news2.near.net!MathWorks.Com!europa.eng.gtefsd.com!howland.reston.ans.net!xlink.net!news.uni-trier.de!ldv01!kordmann
From: kordmann@ldv01.uni-trier.de (Gerhard Kordmann)
Subject: LPC Spectrum, again
Message-ID: <kordmann.778858268@ldv01>
Lines: 32
Sender: root@news.uni-trier.de (System PRIVILEGED Account)
Organization: Universitaet Trier, Deutschland
Date: Tue, 6 Sep 1994 13:31:08 GMT


		Hello everybody,

	I'm sorry to bring up the topic again, but I just cannot
	get an LPC spectrum, no matter what I try. All the dsp
	and speech books show nice pictures of LPC - spectra, that
	look like smoothed fft-powerspectra. But they never say
	how to do it, only how to calculate the LPC coefficients.
	Now, I got 2 different LPC - programs, just to make sure
	the coefficients were OK. They were, at least they were
	identical. So I had 15 LPC coefficients (from the package
	by Patrick Ko, if you know it).
	I prepended a coefficient 0 of value 1.0, as suggested by
	various internetters.
	Then I put these 16 values in the first 16 slots of a
	512 - array, 0 elsewhere. As suggested by many specialists.
	I did an fft, and then by 10*log10(Real[i]*Real[i] + Imag[i]*Imag[i])
	transformed this to a powerspectrum. (Using the first 256)
	Now, according to the numerous internet answers I received,
	I should have the spectrum. But what I get is just a nice
	curve, peaks tending to be in the middle, that has nothing
	at all to do with the powerspectrum of the signal.
	I tweaked here and there, and tried, and read books. But I
	cannot get it to work. I start to despair.
	Doesn't anyone have a piece of working code to do it ? I am
	not a mathematician, so many of the well meant explanations
	were beyond my scope, just a bit of code ? An idea?

		Very confused, Koma



