Mark replied to this mail, but not one the list. Here's a copy of his
answer:
"Hi Gabriel,
I've been thinking about this patch for a while (It is still on my
todo list from when you suggested these ideas a month or so ago).
But I've finally come to the conclusion that it is not a good idea:
I'm worried about the preecho in the frame before the sharp attack.
If you do not window switch, the frame will look something like this:
<----- MDCT window0 ----->
<----- MDCT window1 ----->
<----- MDCT window2 ----->
<MDCT frame0><MDCT frame1><MDCT frame2><MDCT frame3>
If there is a sharp attack at the beginning of frame1, then it will be
important to not use such a large window for frame0. If we switch
to a short block for frame1, then frame0 will be using a 'start' block,
which has a window which does not extend as far into frame1.
Also, to reconstruct the signal, remember that the frames are overlapped
(50% for long blocks) so the output is actually the last 1/2
of frame0 + first 1/2 of frame1. If we switch to short blocks, this
overlap is reduced from 576 samples to 192 samples, so again there is
less contamination of frame0 with frame1."
Gabriel Bouvigne - France
[EMAIL PROTECTED]
icq: 12138873
MP3' Tech: www.mp3tech.org
----- Message d'origine -----
De : Takehiro Tominaga <[EMAIL PROTECTED]>
� : <[EMAIL PROTECTED]>
Envoy� : mercredi 15 d�cembre 1999 11:25
Objet : Re: [MP3 ENCODER] short blocks patch
> uum,,, why no one replies this mail ?
>
> >>>>> "G" == Gabriel Bouvigne <[EMAIL PROTECTED]> writes:
>
> G> Here is a little modification over the block switching to
> G> prevent switching in the case where the big surge is from the
> G> first subblock to the next ones. In this case, long block will
> G> only lead to a post echo of less than 20ms, wich won't be
> G> heard.
>
> I extend this method little.
>
> I think short block is used mainly to reduce the pre/post echo arround
> big surge.
>
> 1.
> A post echo is not noticible than pre echo because of temporal masking
> effect. maybe we can use this effect.
>
> 2.
> When the total energy of passege is small enough, we can't hear each
> of the echo (because of ATH). So we don't need to use short block where
> total energy of block is "enough" small.
>
>
> I don't know where these are right or not, and they need some fine tuning
> of parameters. But it may help quality improvement, I hope.
> ---
> Takehiro TOMINAGA // may the source be with you!
> --
> MP3 ENCODER mailing list ( http://geek.rcc.se/mp3encoder/ )
>
--
MP3 ENCODER mailing list ( http://geek.rcc.se/mp3encoder/ )
