On Mon, 8 Nov 1999, Bill Eldridge wrote:
> I was reading in one source that you can
> 1) convert MP L3 to MP L2 in the digital
> domain without going analog
> 2) that MP L2 (l3?) gain & cross-fades can be
> done in digital domain by adjusting sub-band
> parameters.
>
> Anyone have more info on how this is done/
> how to actually code this?
>
I did not reread other follow-ups, but:
Actually my former grad-student has been working on this;
his progs allowed real-time mixing (including cross fades/in/outs)
i.e. partially decode two ((or more)) mp3/mp2 files, mix them, and
re-encode to mp2. For mp2; it was about 12 times as fast as doing full
decode, mix in time, fully recode.
How to do it (indeed): decode to subbandsample domain, and add
subbandsamples:
a(t)*file1(t) + b(t)*file2(t) etc
Problem: how to do MP2-bitalloc; this is currently without any psychomodel
recalc; so bitalloc is fast, but suboptimal, but still very good.
Does anyone know anything about this (in terms of experiments/scientific
papers about this):
if you mix two streams can you allow more/i.e. worse masking ?
I would think that this could be possible since the brain would try to
track both and hence be somewhat confused, so more errors allowed ?
Additionally the prog also allows fast recoding to lower bitrates e.g. 192
to 128 kbit.
My new grad student is now continuing working on this; currently I feel
that it is maybe still a bit to early to (ask them to) release code; but I
will discuss this with them again.
Maybe you could convince me/them to change our opinion ;)
best regards,
Patrick.
--
MP3 ENCODER mailing list ( http://geek.rcc.se/mp3encoder/ )
