Vorbis dev - Nov 2000 - comparison of beta2 and beta3 encoder quality

Hi,
I encoded some songs into 350 kbit/s with beta2 and beta3 too,
and I don't know why, but it seems the b2 sounds better (makes a
better quality ogg file)...
The b2 contains more the b3 contains less high sounds.
And with the b3, the size of file is smaller with 4-5%...
Do you know this problem, or do you think so, that the sound quality
is good (or better in the b3) ??

Please be carefull with the floating point calculations!
Change them only then, if it seems the new solution/routine
is surely perfect (better or same quality like in original).

Please - if you can - don't use pre-calculated floating point numbers
(i.e.: 0.34563), calculate every numbers in the program (with sin/cos/exp
function and with integer values). Somehow and somewhere the pre-calculated
numbers were also calculated...

Please use a 'typedef' for the floating point precision too (i.e.
typedef
float ogg_float_t), but I would be happy if you would use 'bouble' by
default...
Or at least, the possibility of the more precision and more clear sound.

So, the Ogg Vorbis is very good, maybe with these modifications it could
be perfect too.

incerely
Attila Padar
author of Mpxplay (ogg support under construction)

--- >8 ----
List archives:  http://www.xiph.org/archives/
Ogg project homepage: http://www.xiph.org/ogg/
To unsubscribe from this list, send a message to
'vorbis-dev-request@xiph.org'
containing only the word 'unsubscribe' in the body.  No subject is
needed.
Unsubscribe messages sent to the list will be ignored/filtered.

Attila Padar wrote:
> 
> Hi,
> I encoded some songs into 350 kbit/s with beta2 and beta3 too,
> and I don't know why, but it seems the b2 sounds better (makes a
> better quality ogg file)...
On some music, beta2 sounds better (imho); on most, beta3 does. Note
I'm encoding at lower bitrates than you are, so maybe beta2 is better
on most for 350kbps.
> The b2 contains more the b3 contains less high sounds.
> And with the b3, the size of file is smaller with 4-5%...
That's normal: Vorbis doesn't try to get the bitrate you request,
it's
just an indication for how exactly it does the coding (which psycho
acoustic parameters to use).
> Do you know this problem, or do you think so, that the sound quality
> is good (or better in the b3) ??
Better, most of the time.
> Please be carefull with the floating point calculations!
Yes, they will attack when you least expect-em to ;-)
> Change them only then, if it seems the new solution/routine
> is surely perfect (better or same quality like in original).
What makes you think this is not what happened between beta2 and
beta3?
> Please - if you can - don't use pre-calculated floating point numbers
> (i.e.: 0.34563), calculate every numbers in the program (with sin/cos/exp
> function and with integer values). Somehow and somewhere the pre-calculated
.034563 is 17 (and a half) bit of precision; on more digit and it's the best
you'll ever need. Input and output is only 16 bit pcm; that's 15 bits of
precision. Having 20 bits for intermediate calculations is all you need,
provided your algorithms are numerically stable. If not, you lose anyway.
> numbers were also calculated...
> 
> Please use a 'typedef' for the floating point precision too (i.e.
typedef
> float ogg_float_t), but I would be happy if you would use 'bouble'
by
> default...
What's a bouble? Or do you mean double? The cahnge from double to float was
not exactly entirely accidental, or not well-thought...
> Or at least, the possibility of the more precision and more clear sound.
You can have better precision if you like, but it won't give you better
sound.
> So, the Ogg Vorbis is very good, maybe with these modifications it could
> be perfect too.
It will never be perfect, sorry to burst your bubble, but real life
stinks.

Ciao,

Segher
--- >8 ----
List archives:  http://www.xiph.org/archives/
Ogg project homepage: http://www.xiph.org/ogg/
To unsubscribe from this list, send a message to
'vorbis-dev-request@xiph.org'
containing only the word 'unsubscribe' in the body.  No subject is
needed.
Unsubscribe messages sent to the list will be ignored/filtered.

Segher Boessenkool (segher@wanadoo.nl) wrote :
>    Attila Padar wrote:
[ snip ]
>    > Please - if you can - don't use pre-calculated floating point
numbers
>    > (i.e.: 0.34563), calculate every numbers in the program (with
sin/cos/exp
>    > function and with integer values). Somehow and somewhere the
pre-calculated
> 
>    .034563 is 17 (and a half) bit of precision; on more digit and it's
the best
>    you'll ever need. Input and output is only 16 bit pcm; that's 15
bits of
>    precision. Having 20 bits for intermediate calculations is all you need,
>    provided your algorithms are numerically stable. If not, you lose
anyway.
Will vorbis never support 24 bit ( or more ) input and/or output ?
      
>    > numbers were also calculated... 

-- 
David Balazic
--------------
"Be excellent to each other." - Bill & Ted
- - - - - - - - - - - - - - - - - - - - - -
--- >8 ----
List archives:  http://www.xiph.org/archives/
Ogg project homepage: http://www.xiph.org/ogg/
To unsubscribe from this list, send a message to
'vorbis-dev-request@xiph.org'
containing only the word 'unsubscribe' in the body.  No subject is
needed.
Unsubscribe messages sent to the list will be ignored/filtered.