[CELT-dev] CELT 0.6.0 released, submitted as IETF draft

[Gregory Maxwell]

On Mon, Jul 6, 2009 at 5:40 PM, John Ridges<jridges at masque.com> wrote:
> Congrats JM. I've been looking forward to 0.6.0 for a while now, and
> I've just integrated it into my app and it works great. I have just one
> question though. I realize that the VBR implementation is brand-new and
> isn't finished yet, but I would like to use it and was wondering if you
> could give us all some idea of how you envision the VBR eventually
> working (in other words, how the VBR parameter would affect the
> bit-rate). Right now it appears to give me about half the bit-rate I ask
> for when encoding silence, bit will sometimes increase up to around the
> full bit-rate on occasion depending on the audio.

As far as silence goes, if you're sending in true digital silence we
should eventually have code which detects that and emits zero sized
packets.  If you're sending in very quiet analog noise things are
quite a bit harder because we can't be sure how loud the far end will
have turned up their volume knob.

The volume control knob problem is addressed in vorbis by the encoder
running an envelope follower that works on the assumption that the
loudest sound in the last X seconds will be considered to be Y dB.
We'll probably have something like that in CELT anyways, and perhaps
expose an API to set hard limits on the follower, so that your quiet
near-silence would still get treated as silence.

We're just not quite there yet.