Google Funds Ogg Theora For Mobile

An anonymous reader writes "Google has decided to fund the development of Theora optimized for ARM processors. The article on the Open Source at Google blog notes the importance of having a universal baseline video codec for the Web: 'What is clear though, is that we need a baseline to work from — one standard format that (if all else fails) everything can fall back to. This doesn't need to be the most complex format, or the most advertised format, or even the format with the most companies involved in its creation. All it needs to do is to be available, everywhere. The codec in the frame for this is Ogg Theora, a spin off of the VP3 codec released into the wild by On2 a couple of years ago.'"

Comments

Paging Chris DiBona

[David Gerard]

Chris DiBona of the Google open source group claimed that "If [youtube] were to switch to theora and maintain even a semblance of the current youtube quality it would take up most available bandwidth across the Internet."

This was shown to be false.

Mr DiBona then mysteriously vanished without trace.

Could he please manifest and either (a) support his claims or (b) concede his error?

Thanks ever so much.

Re:Paging Chris DiBona

[Graymalkin]

The Xiph's group rebuttal page does nothing to show Chris DiBona's contention was false. As I have said before, through either ignorance or malice the Xiph guys dropped the ball on their comparison.

1. Their larger Theora video has an audio track that's about 64kbps. The H264 video from YouTube has a 128kbps audio track (the numbers are rough since they're VBR tracks). This means for every second of video the Theora video has an extra 64kbps to throw at the video. While 64kbps might not sound like much that's 13% of the file's total bitrate. This gives the Theora track a 13% data rate advantage over YouTube's video. Every objective test I've ever seen has gauged AAC and Vorbis to have roughly equivalent audio quality at the same bitrate. If they want to make an actual comparison they would need to use a 128kbps Vorbis audio track.

2. The Ogg file format really sucks for streaming over the internet. The Ogg container tries to be too general of a format when it's only being used to represent time based media. FFMPEG developer Mans has a lot to say about the container format. Thanks to sample and chunk tables in the MPEG-4 format seeks are really efficient over the network since the header gives you an index to all of the samples in the file. A single HTTP request or file seek is needed to seek to a particular time in the file, even if the full file hasn't been downloaded yet. For services like YouTube and Vimeo, especially in context of mobile connections, Ogg's inefficiency is a real detriment.

3. MPEG-4 files with H.264/AAC tracks can be handled by the Flash plug-in as well as natively in browsers. YouTube and Vimeo and others can encode a single version of a file and serve it up to older browsers using Flash and newer browsers using the HTML5 video tag. If Ogg is added as an option that is another step in your decision tree. For individual requests this extra logic might be trivial but when you're handling millions of requests per hour this really adds up.

I'm not defending any hyperbole Chris DiBona was spouting off about the internet grinding to a halt but Ogg and Theora are simply not optimal for a "baseline" media format. It's only real feature is the fact it is open source and doesn't require a license. This isn't the most useful feature in today's world because all of the mobile devices that would be served Theora files already have licenses for MPEG-4. Tens to hundreds of millions of phones already support MPEG-4. They're using MPEG-4 to do send video over MMS and e-mail and for watching video on the web. Theora improve any of those experiences.

Re:Dirac

[TheRaven64]

CPU load. Theora is based on VP3, which is old. It was open sourced in 2004, but VP3 first shipped in 2000. Back in 2000, I had a 450MHz K6-2, and a lot of people I knew had slower machines. Now, a typical handheld is faster than that machine. Theora, like VP3, relies a lot on postprocessing passes for quality. This has the advantage that you can just not bother on slower machines, and get a slightly worse picture but with a lower CPU requirement.

Dirac, in contrast, needs at least a 2GHz CPU to play back. It's patent free and looks great, but the CPU load is huge. There have been efforts to offload a lot of it onto the GPU, which is nice for the desktop but doesn't help older machines and handhelds (except the latest generation). The BBC is working with vendors to get Dirac implemented in hardware, but it won't be ubiquitous for quite a few years.

Dirac also doesn't perform as well as Theora at low bitrates. This is very important for web streaming. Dirac is great for situations where bandwidth and CPU power are plentiful, but Theora makes more sense as a lowest common denominator solution. Ideally, you'd see both supported; Dirac for high quality, Theora for fallback.

Re:WHATWG: The worst thing to happen to the Web.

[Anonymous Coward]

Meh, getting 503s trying to log in. Sorry for the A/C Post.

XHTML was interesting and lovely, and no one gave a shit. Ideology loses to practicality in almost every case until ideology is reformed to conform to reality.

I think you'll find that if you look at HTML5, there's not a lot of presentational bits in it. Most of that is still reserved for CSS.

You'll also find that the cases where things are defined at least gives the web a unified a way to handle real web pages that exist *today*. Right now, a new browser would have to reverse engineer what Chrome, FF, IE and friends did in order to know how to render the web. HTML5 at least identifies the reality that exists.

You note that JS is being used to do things it shouldn't. On what grounds? Who are you to tell what should and shouldn't be done with a language and in a given environment? The practical fact is that folks *are* doing amazing things with JS. If you don't like the language, that's your problem. If you don't want it on your computer, don't use those websites. JS *does* lots of things today, and there's no reason to limit it artificially. You want something better out there? Come up with a solution and push it.

Your final comment notes that web developers aren't interested in quality and technical superiority. You're right. Why should they? What they care about is getting a product out. You're asking them to solve problems that they don't have.

Tks,
Jeff Bailey
(an employee of Google, not speaking for Google at all)

Beyond awesome!

[KingSkippus]

This is beyond awesome, it's a game-changer. Google is one of those rare companies that singularly has the power to move markets, and it is revolutionary to see it do so in favor of consumers as it has. I understand the reasons why it has preferred H.264 over Theora, but it is really nice to see that it also understands the reasons why we should be preferring an open format instead. It's especially nice in an age of companies wanting to lock everything down and be the gatekeeper to everything, the major player in technology is pushing yet again to open things up.

Sometimes I think that Google is about the only company that "gets it." They understand that more people using the Internet translates to more money in their pocket. Even if those people are not using Google's services directly, they are increasing the market such that collectively, it has more opportunity, which in turn translates into more $$$. They seem to not really care if other people are making more money as well, which really separates them in my mind from other companies, who are of the "it is not enough that I succeed, but everyone else must fail" mentality.

Anyway, back to the topic at hand, one reason I've seen people regurgitate in why H.264 is the right way to go is because it is supported on hardware. Congratulations to Google on working to negate that argument.

Re:OGG newbie question

[nxtw]

How does OGG compare to MPEG4?

Theora is perhaps better than H.263 and MPEG-2 (from the mid 90s), but does not come close to H.264/MPEG-4 AVC or VC-1. (The frozen Theora bitstream format is lacking many features found in H.264 and VC-1.) Results might be similar to H.263+/MPEG-4 ASP.

The Ogg container also has some documented flaws.

Note that there are many sites which perform misleading or flawed comparisons of the two; for example, they might compare the result from YouTube's H.264 encoder with a lossy source (which optimizes for encoding speed) to a locally ran Theora encode with a lossless source.

Since OS X 10.6 and Windows 7 come with H.264 decoding, and Windows 7 supports H.264 hardware decoding with compatible hardware from any source, I recommend sticking with H.264. (OS X 10.6's H.264 hardware decoding support appears to be limited to videos played in QuickTime X from MPEG4 or QuickTime container files on systems with nVidia 9400M GPUs or newer, even though Macs with capable GPUs started appearing in 2007.)