Previous Entry Add to Memories Share Next Entry

We have met the CELT and they is us.

A quick post to mention that I've just put a new demo page up introducing a new audio codec named CELT. It's Xiph's first next-generation audio codec now emerging from the development pipeline, intended for high fidelity use at ultra-low latency.

Follow the above link to a ton more about it. As usual, the demo's got plenty of... well... demo, in addition to eye candy and hard information.

Buried in that larger announcement (about a much larger project) is a mention in passing of Squishyball, a cute little app I wrote last month to do fast/easy/convenient comparison between groups of similar samples. It also performs more rigorous A/B, A/B/X and X/X/Y double-blind testing. We'd all been limping along with harder to use apps that caused us to avoid or put off careful listening tests as part of day-to-day codec development. Hopefully Squishyball fixes that.

"next-generation codecs (Ghost and CELT) as successors to Vorbis."
Does that mean Vorbis is done for? Is Ghost supposed to be the new Vorbis?
"(see the 'Thusnelda' and 'Ptalarbvorm' encoders)"
You said Ptalarbvorm is part of the main Theora branch, does that mean Ptalarbvorm is finalised?

So is Ghost next, after the bitstream is frozen, or Speex?
(Frozen) (Thread)

"The Vorbis codec is more than halfway through its approximate intended lifetime of 20 years or so,"

By this figuring we have about 9 years to finish ghost as a replacement for Vorbis. I'm sure Monty has dreams of finishing it faster than that, but it always takes longer than you expect.

CELT primarily fills in areas where Vorbis (or regular AAC, or MP3) can't even be used at all— low latency. Though it does out-do Vorbis significantly for some use cases even when latency isn't a consideration (low bitrate speech, for example; or low memory environments).

The combination of CELT and Silk going on in the IETF (Opus) will obsolete Speex, I expect. (With the exception of the Speex installed base, Opus will be pretty much strictly superior to Speex right out of the gate).

Ptalarbvorm (libtheora 1.2) isn't finalized, it's the mainline development for libtheora right now. CELT is currently competing with libtheora for development resources, especially with the CELT bitstream freeze upcoming.
(Frozen) (Parent) (Thread)

Hi, Greg; since you are a main commiter of Ptalarbvorm, how far along would you say it is? (e.g. alpha, beta)

"By this figuring we have about 9 years to finish ghost as a replacement for Vorbis."
Ok, so Ghost is a replacement for Vorbis.

"On March 3, 2009 Skype Limited announced that the SILK codec will be available soon under a royalty free license to third-party software and hardware developers" -Wikipedia
So Speex will stop being developed when SILK is royalty free?
(Frozen) (Parent) (Thread)

Ptalarbvorm is "alpha" right now, I guess. I think that right now we're just doing a last round of performance and quality improvements in the encoder before it's released. The release schedule is really up to Tim. To the best of my ability to determine the code in SVN right now is perfectly fine from a software quality perspective and /could/ be cut as a release ~today.

Our release plans have slipped a bit, in part because I've failed to finish some one of the sub-projects (The temporal RDO, oops, sorry), and because of competing demands from CELT (Tim and I have been the only active committers through most of this Theora cycle, and we're both active on CELT.

Speex hasn't been under _active_ development for a while because Jean-Marc (primary developer of Speex) has been focused on CELT. SILK is now going to become royalty free as part of the process of merging it with CELT and standardizing it in the IETF.

(Frozen) (Parent) (Thread)

"SILK is now going to become royalty free as part of the process of merging it with CELT and standardizing it in the IETF."
There will still be CELT, standalone, right? Considering SILK+CELT would be best for voice. There are still instances where people would use it for things other than talking, e.g. online concerts (as mentioned in the demo page).
(Frozen) (Parent) (Thread)

You are viewing xiphmont