[FFmpeg-devel] Flac encoder and MPEG4 ALS

Kostya kostya.shishkov
Sun Mar 30 17:44:40 CEST 2008


On Sun, Mar 30, 2008 at 08:53:12PM +0000, Jai Menon wrote:
> On Sunday 30 March 2008 13:55:49 Kostya wrote:
> > On Sun, Mar 30, 2008 at 09:44:51AM -0400, Justin Ruggles wrote:
> > > Kostya wrote:
> >
> > [...]
> >
> > > > I've looked at it once. Seems the most features are for floating-point
> > > > mode, baseline decoder should be comparatively easy. Especially if you
> > > > are the person who writes encoder to it ;)
> > >
> > > It has 2 entropy coders, plus another for floating-point.  It has 2 ways
> > > of doing multi-channel.  There are also several prediction methods.  The
> > > spec does not specifically say what is "baseline" vs. extended, but it
> > > would be simple enough to break down the task.
> >
> > IIRC, both entropy coders are VLC - classical Shannon-Fano scheme and
> > classical Golomb codes. Prediction is also standard LPC and RLS, the latter
> > is quite rare in codecs though (MLP, APE, TTA ?).
>  and iirc golomb rice codes are used only for the tail end residuals, those in 
> center are encoded using block gilbert-moore codes........but i may be wrong

Just to clarify - Rice codes are one case of Golomb codes and yes, they are
used for large values in ALS.

Block Gilbert-Moore codes are another name for Shannon-Fano codes and
respect Shannon more than that RealNetworks engineer who suggested that
coding scheme and name for ALS.
 
> Regards
> Jai Menon
> <realityman at gmx.net>




More information about the ffmpeg-devel mailing list