I’ve been involved in online conversations about streaming media formats (mp3, AAC, Ogg Vorbis, Real, Windows Media and so on…), and along the way the inventor of the Ogg format has become peripherally involved. I thought it might be worth noting some of his observations that were posted in response to some public mailing list questions:
> Most say that a 48kb/s AACplus sounds better than a 128kb/s MP3
That’s claiming a bit much. Even against the original Blade (MP3 encoder) this would require not paying much attention. What’s true is that the artifacts sound very different, so if you’re listening for mp3-like high-end problems, you probably won’t hear them.
> which would put Ogg Vorbis at around 96kb/s IMO.
Not really. 64kbps Ogg is still a little better than the 48kbps AAC+SBR we’re talking about.
That’s not to say the low bitrate AAC+SBR isn’t very good; as a new codec, it’s quite an improvement to classic AAC at low bitrates. But many of the press-release claims are (understandibly) pushing the envelope of what is really believable.
Also, it’s absolutely true that at the lowest bitrates (32/48kbps) AAC+SBR currently has a decent edge over Vorbis low bitrate. Everyone is still moving forward and the lead changes back and forth. The low bitrate race is pretty much down to new AAC and Ogg these days. At low bitrate, AAC is ahead, at mid-bitrate Vorbis is still leading.
> Is there anything that can be done to bring Ogg Vorbis up to this type of
> quality in the future, or is it about as good as it’s ever going to be?Of course. Development of these things tends to move forward in bursts. I’ve done little Vorbis research work since 1.0 as there have been other things to do. I’ll get back to it if only because it’s what I’m best at and audio codec research feels good ;-)
> I would rather use patent free and open codecs, but this type of bitrate
> saving, particularly for streaming, cannot be ignored, and I’m concerned
> that this will slow the uptake of Ogg Vorbis and may reverse it’s
> popularity in time.The bitrate savings isn’t as big as you fear and we’ll take the lead back again. Everyone is improving. Who would have ever thought back in ’98 that LAME would get as good as it is today? I’m curiously surprised that AAC hasn’t moved forward farther than it has…
Monty
—————————————
And here are some more responses from Monty> Now if that is true, technically it would be possible to use the same
> technics (SBR and PS) added on Vorbis OGG.No, because the specific SBR extensions are part of the MPEG patent pool. As it is, the current/future direction of Vorbis is to move away from a pure transform-domain encoding anyway. SBR would still be applicable, but some of the same mechanics would be built into the rep model.
SBR’s fundamental basis is ‘each octave is closely related to the octave below it’. It then implements something akin to an ‘exciter’ that represents most of the upper part of the spectrum as a parametric set of harmonic duplications of the lower octaves.
The basic idea is both sound and one that’s been underexploited (upper octaves are often harmonic duplications of lower octaves). One can take advantage of that without using SBR, and that’s the direction I’m going. But there’s more than the harmonic structure here; if strong tones (the harmonic structure) of the audio is subtracted out, and you’re only looking at the leftover ‘noise’ in time, the subband
correlation is very strong in time as well. I don’t know if AAC(classic)’s Temporal Noise Shaping is also exploited by the SBR extensions in any way, but I certainly plan to exploit that fact in future Vorbis.Monty
—————————————
> But, at its best, with an ‘average’ bit rate of 128kbps, mp3 should easily beat 48kbps aacplus. (Even at 128kbps cbr, I’d expect it to beat aacplus!)
Yes, add me as a ‘me too’ :-)
A decent 128kbps encoder should hand AAC+SBR its ass on a platter. Respectable 128kbps mp3 encoders are very good today. “Something is wrong”.
> But at lower ‘streaming radio’ rates, no, Vorbis isn’t the best. It
> wasn’t tuned / designed for those rates.
> Roberto’s 32k rate listening test didn’t show Vorbis in a good light… I
> doubt a 48k test would be much different.
>
> Also, I faintly remember there were some comments after the test.
>
> I think that maybe they used some poor settings or some resample
> issues or something. The result was that vorbis might not have been at its best.Speculation aside, I would not have expected Vorbis to fare well at 32kbps. It simply wasn’t tuned/designed for it. Vorbis (as it is now) was designed to scale down to ~64kbs at a time when mp3 wasn’t yet respectable below 192kbps. Today’s AAC is not actually the same codec as the AAC everyone remembers from three years ago; they’re different, related but incompatable codecs taped together in a new
package under the original name. No worries, we will update Vorbis similarly and keep the same name so no one knows the difference :-)Monty
Watch this space…
