Forums

LPC in physiological sense

Started by HyeeWang July 24, 2009
We can attain the all pole system function of speech from linear
prediction(LP). But how to correspond it with the digital model in
physiological sense?
Where can i download such literature discussing the relation between
them in detail?

questions: 1. what is the counterpart of the lpc all pole model in
physiological  sense?
         2.  what is the counterpart of the input of lpc all pole
model then?

HyeeWang
The all pole model is as follows.
 H(z) = G/(1-sum(ak*z.^k))
where,  a is the lpc coefficients.

The physiological  digital model is : exitation system + vocal tract
system +  radiation system.
On Jul 24, 2:36&#2013266080;pm, HyeeWang <hyeew...@gmail.com> wrote:
> The all pole model is as follows. > &#2013266080;H(z) = G/(1-sum(ak*z.^k)) > where, &#2013266080;a is the lpc coefficients. > > The physiological &#2013266080;digital model is : exitation system + vocal tract > system + &#2013266080;radiation system.
H(z) = G/(1-sum(ak*z.^ -k)) . It is -k,not k.

HyeeWang wrote:
> On Jul 24, 2:36 pm, HyeeWang <hyeew...@gmail.com> wrote: > >>The all pole model is as follows. >> H(z) = G/(1-sum(ak*z.^k)) >>where, a is the lpc coefficients. >> >>The physiological digital model is : exitation system + vocal tract >>system + radiation system. > > > H(z) = G/(1-sum(ak*z.^ -k)) . > > > It is -k,not k.
On Jul 24, 12:40=A0am, HyeeWang <hyeew...@gmail.com> wrote:
> On Jul 24, 2:36=A0pm, HyeeWang <hyeew...@gmail.com> wrote: > > > The all pole model is as follows. > > =A0H(z) =3D G/(1-sum(ak*z.^k)) > > where, =A0a is the lpc coefficients. > > > The physiological =A0digital model is : exitation system + vocal tract > > system + =A0radiation system. > > =A0H(z) =3D G/(1-sum(ak*z.^ -k)) =A0 . > > It is -k,not k.
This is a specific system. But in DSP point of view, you got to find feed back loop in the counterpart system.
HyeeWang wrote:

> 1. what is the counterpart of the lpc all pole model > in physiological sense?
None. The LP source-filter model is only phenomenological and cannot be decomposed into models of the bodily contributors to voice production. In particular the vocal folds do not produce a pure impulse train, so a first step toward physiological correspondence could be to record the actual excitation with a thorax microphone and deconvolve the sound signal with that for the vocal tract response, though this still leaves both parts as black boxes. For an example of deeper modeling see: Kob et al., Time-domain model of the singing voice, DAFx'99 ftp://ftp.funet.fi/pub/sci/audio/dafx/1999/www.tele.ntnu.no/akustikk/ meetings/DAFx99/kob.pdf Martin -- Quidquid latine scriptum est, altum videtur.