Talk:Linear predictive coding

Latest comment: 1 year ago by Chris2crawford in topic Underlying technique

Underlying technique

edit

This page seems to cover the major application of LPC, without considering the underlying technique very well. There is a good introduction in Numerical Recipes. — Preceding unsigned comment added by DoctorRad (talkcontribs) 16:20, 3 March 2018 (UTC)Reply

I agree! Chris2crawford (talk) 15:52, 1 August 2023 (UTC)Reply

Kelly-Lochbaum

edit

Could someone please add equivalence to the Kelly-Lochbaum vocal tract model? At least on the history section. I'll do it in a while if I have the time.... Oyd11 23:02, 18 June 2006 (UTC)Reply

Lossless tube model

edit

I wanted to ask if it makes sense in your opinion, to add an article "lossless tube model" and to link it to the first sentence of the "Motivation" subsection. I think such an article would be useful, since Rabiner and Schafer use in their book "Digital Processing of Speech Signals" about 16 pages to cover the theory of a lossless tube model. --Bjoern.thalheim 12:36, 5 August 2006 (UTC)Reply

Perceptual Linear Prediction

edit

PLP mentions "Perceptual Linear Prediction", but those words are currently unlinked. Should it link here? (is "linear predictive coding" the same as "perceptual linear prediction" ?) --68.0.120.35 17:15, 28 January 2007 (UTC)Reply

Perceptual Linear Prediction should probably have its own page which references this one - while it's based on the same idea as LPC (i.e. fitting an all-pole model to a short-time digital speech signal), the algorithm is different from the standard ones used in basic LPC analysis. DavidHugginsDaines 14:17, 16 April 2007 (UTC)Reply

Merge discussion

edit

Merge discussion is here

Use in computers

edit

The Speech Synthesizer for the Texas Instruments 99-4 and 99-4/A Home Computer was able to use LPC for generating natural sounding speech. Notable uses of it were the games "Parsec" and "Moon Mine".

The IBM PCjr used the same Texas Instruments speech synthesizer chip in its optional speech sidecar as TI used for the Home Computer's speech option, though IBM's implementation was not as versatile and very few PCjr programs made use of it. (The PCjr also used the same TI Complex Sound Generator chip as TI's computers, as did the Radio Shack Tandy 1000 computer series.)

Other computers and game consoles like the Magnavox Odyssey^2 and Mattel Intellivision used pre-digitized audio for speech, or played back recorded audio from a cassette tape.

LPC coefficient representations

edit

There is an error in this section. LCP filters could never be unstable so i was thinking of rewriting it as:

"Transmission of the filter coefficients directly (see linear prediction for definition of coefficients) is undesirable, since they are very sensitive to errors, i.e., a very small error can distort the whole spectrum. However, LPC filters are always stable as they belog to the Finite Impulse Response (FIR) filter class, which is one of their greatest advantages."

but then the whole section should be rewriten and i dont feel confident about it. I dont have enough knowledge about advanced representations (LAR, LPC and ParCor) to do so. Could someone take the job? — Preceding unsigned comment added by Karagiosis (talkcontribs) 15:59, 5 October 2011 (UTC)Reply

Well, the above remark is just wrong. You could check here: http://en.wikipedia.org/wiki/Infinite_impulse_response LPC produces an all-pole model, meaning a polynomial in z in the denominator, and if the roots of that polynomial lie outside the unit circle in the z-plane (equivalent to lying on the right-hand side of the s-plane, meaning s with a positive exponent) then the filter output grows with time and that growth is unbounded.

But my real problem with this whole article is that it doesn't even mention Manfred Schroeder. Prof Schroedey claims here http://www.physik3.gwdg.de/~mrs/#ref that he invented linear prediction at Bell Labs, along with Bishnu Atal. I never met Prof Schroeder, but I was working in speech research in 1967 and met Dr Atal. Their 1967 Bell Labs article, in the reference by Schroeder that I just listed, was certainly where I and everyone around me first heard of linear prediction -- though we had a US East Coast bias and the work of Saito and Itakura soon also became significant. In particular, the Itakura method of using lattice filters was a breakthrough in implementing the LPC process. 2.25.176.128 (talk) 13:29, 18 November 2011 (UTC) Richard WrightReply

OggVorbis code from xiph.org uses an autocorrelation LPC coefficient generation algorithm that was invented by N. Levinson in 1947, modified by J. Durbin in 1959. Seems a lot of history was omitted herein.