[flac-dev] Confusion about linear prediction within flac

Timothy B. Terriberry tterribe at xiph.org
Sun Sep 9 20:01:41 UTC 2018

Robin Patrick Decker wrote:
> I would really appriciate an explanation or information on a good
> resource to learn more about how the prediction coefficients are solved
> for.

The Wikipedia page on this subject is not terrible:

The very high-level answer is that if want to choose your coefficients 
to minimize the mean squared error of the prediction, then you get a 
least-squares problem where the matrix you're inverting is just the 
auto-correlation matrix of your signal (the Yule-Walker equations). The 
discrete Fourier transform is just a computationally efficient means of 
computing the auto-correlation, at least if your order is sufficiently high.

> Once the lpc coefficients have been solved, as far as I understand you
> must also store part of original signal with length equal to the
> prediction order since you need the previous samples to predict the
> next sample of which only the residual is known. For the next values
> the  residual is used to retrieve the original value which is fed into
> the predction model to further reconstruct the time series. Is this
> correct?


More information about the flac-dev mailing list