[CELT-dev] CELT Test case failures

[Gregory Maxwell]

On Fri, Aug 5, 2011 at 4:08 PM, Andrew Lentvorski <bsder at allcaps.org> wrote:
> So, all of the issues with using Opus instead of CELT have been resolved?

Many but not all.

> It's now a part of the main Vorbis repository?

No, and it never will be. It's in the official opus repository, where is
been for a few months since we stopped using the split sub-repositories.

> There's an official
> snapshot?

Yes, and there has been.

> There is now documentation?

For the library API, not much.

Those cycles have been spent on writing a formal specification
of the format.

It's at least observably enough for someone who has never used
the API before to begin using it, as we've heard from some people
who recently implemented support.

The headers document the parameters.

The test app shows some usage. There is a opus-tools package currently
being written (available on git.xiph.org already) which will ultimately
be neater and more useful.

> There is now example code that works out of the box and doesn't require
> the knowledge of arcane switches to make work or at least emits sane
> error messages?

It should. Bug reports and patches are generally accepted.

> There is now quantification of how the CPU usage profile has changed by
> switching from CELT by itself to the hybrid codec?

Sure but my records don't reflect you asking about this before, if so
I missed it. Sorry about that.

If you don't want to use hybrid mode, don't use it.

In general, the LPC (silk) encoder is much slower than CELT in the full
complexity mode, and the decoder is much faster. Because of this
from a decode side the complexity of hybrid is basically
negligible compared to CELT alone.

> I also note that SILK
> requires a patent license.  I presume this was discussed on a Vorbis
> list somewhere, could someone give me a pointer so that I can avoid
> rehashing that off-topic discussion on this list.

https://datatracker.ietf.org/ipr/1602/

As of a week-ish ago the skype license even applies to the
pre-approval draft; the broadcom and Xiph licenses do not which
yet is why there hasn't been an announcement with respect to
that.  I wouldn't mind seeing commentary on what you thought of
that license.


I regret not communicating on the list more often— unfortunately there
are so many balls in the air and so little time. Opus is a volunteer
project for me (and, has for the other developers too much of the time)
it's easily been more than a thousand hours of work, and thousands of
dollars in costs (travel for the standardization effort, lost work,
obtaining patent documents, etc).

Much of the work has been fun, but much of it especially more recently,
has been every difficult and unenjoyable— intense politics, legal
wrangling, and coping with ungrateful jerks on the mailing lists.

I do this work because I want the world to have access to a free
format with the features we've created here. Of course, I want the
package to be as easily useful as possible— with excellent
documentation and a mature build system. But ultimately these
things take a back seat to the format itself.

These things will be improved in time— faster if people help out,
if thats not your style feel free to write your own implementation.
We've already done the hard part of designing the format.