[vorbis-dev] Re: [MP3 ENCODER] long message on absolute threshold of hearing (ATH)

[Greg Maxwell]

On Mon, 6 Sep 1999, Mark Taylor wrote:

> Robert recent coded up (in LAME) the formula for the ATH.  I would
> like to switch the LAME psycho-acoustics (gpsycho) over to using
> formulas for all quantities like this instead of the ISO MP3 tables.
[snip]

The 'black box' tables the ISO mp3 encoder uses have always troubled me.

Reasons why FhG might have used magic tables insted putting in the actual
algs:

1) Table design fits better for implimentation in hardware/tables faster
        in software too.
2) Decoder is only supposed to handle a limited number of settings, so
        nothing more then tables is really needed.
3) FhG wanted to keep their generation details secret so that their
        encoders would always have a leg up on the competation.
4) FhG not only wanted to keep the generation details a secret but wanted
        to include substandard tables in the reference design.

To me 1 (and 2 somewhat) seem unlikely because the refrence design shows
no other evidence of performance tuning. When looking at the number of
bugs we've already found in the ISO code, 3,4 don't seem so unlikely esp
considering that the reference AAC encoder includes a dummy psyco model.

Does anyone have an old old old copy of l3enc? It would be *very*
intresting to see if it has the bugs that are in the ISO encoder.

If it is likely that the reference tables are 'bad by design' then we
should be making a strong effort to replace them, preferably with nice
little modular replacements that can be understood, used in other
encoders, and twiddled with.

--- >8 ----
List archives:  http://www.xiph.org/archives/
Ogg project homepage: http://www.xiph.org/ogg/