[Lame-dev] block type decision / speed

[Gabriel B. <bou...@mp...>]

Right now, lame is computing masking for both long and short blocks. It's
because if next_block_type!=current_block_type then the current_block_type
has to be changed.

As an example if current is long, then if next is short then current should
be changed to start.

But it seems that it would also be possible to only compute masking for the
proper definitive block type. It should be possible to check for attack in
the next block before encoding the current one, and then compute masking and
quantize only the definitive current block type. This would probably lead to
a good speedup in encoding.

Now the question is: do you think there could be some technical problems
associated with this method (except of course the fact that this would
probably need a lot of changes into libmp3lame)?


Regards,

----
Gabriel Bouvigne
www.mp3-tech.org
personal page: http://gabriel.mp3-tech.org