Tonight's commit is a round of psychoacoustic improvements and bugfixes.
(Basically, a week spent with plotting tools making sure each and every element
of the process was functioning as expected. Found some nasty little tuning
mistakes in there). Things are currently tuned to approximate 128kbps.
If you find any samples that don't sound good at this point, let me know.
What's next, what's left:
There's an MDCT bug I hadn't noticed before. The code, as committed,
doesn't
trip it (the blocksizes in use right now are the ones that work correctly).
I'll catch that in a bit.
I backed out time domain envelope clamping for now. When it works, it lets you
practically get away with murder in a pre-echo sense. There were a number
of samples (like castanets) where I could use a single blocksize of 2048 and it
sounded flawless. There are more, unfortunately, where clamping made things
much worse... Until I figure out a better way to do it, I've disabled it
entirely (the hooks are still there). Like continuous balance, I'll have to
come back to it.
Next thing I'll begin on: The codebook module. It's about time Vorbis
starts
writing real (not toy) streams.
Monty
--- >8 ----
List archives: http://www.xiph.org/archives/
Ogg project homepage: http://www.xiph.org/ogg/
