[vorbis-dev] [Jan.Tangring@et.se: spectral band replication]

[Robert Voigt]

> ----- Forwarded message from Jan.Tangring at et.se -----
> So, unique, he says. But obvioulsy he is partial on that, cause it is
> his own patents he is talking about. So I would very much like to
> have a second opion on that, and I don't know ehere else to turn to
> but to this open community.
>
> Do you (or the ogg vorbis community) understand how SBR works? There is
> info on http://www.codingtechnologies.de/technology/sbr.htm
> bnut I am guessing this infop is not enough to explain the technology.
>
> Is SBR really a unique technology or is it just part of some kind of
> patent maneuvering strategy?

I haven't heard about SBR before. After reading that webpage I can say the 
following:

At low bitrates the higher frequencies are cut off. With SBR, these 
frequencies are added again in the decoder, using some hints from the 
encoder. These hints can only use very little bits, otherwise you could just 
encode the high frequencies as usual. I think these hints just say "this 
sounds like voice" or "this sounds like a wind instrument", and the decoder 
reconstructs the spectrum accordingly. This way you make assumptions on the 
source, which reminds me of vocoders. They achieve very low bitrates for 
voice, because they make the assumption that the source is human voice, and 
then generate the whole spectrum from a few bits. Audio coders like mp3 or 
Vorbis do not make assumptions on the source, if they did they couldn't 
reproduce all kinds of sounds properly. 
But SBR may make sense for very low bitrates. If you have to decide if you 
want to have an arm or a leg cut off, you may want to say: "I'd rather have 
my arm cut off, because I need the leg (lower frequencies) for walking and 
it's easier to replace the arm with a prosthesis (extrapolate the higher 
frequencies)." But what if you're an alien from outer space and no one knows 
how your arm looked and how it worked (as a real alien you had only one 
arm!), and they can only give you a prosthesis of a human arm, which doesn't 
fit? The same way SBR will fail if there's an unusual sound the developers 
didn't take into account. It will sound like shit then. But for all the 
sounds they did take into account SBR may actually have an advantage over 
just cutting the high frequencies off or coding them very imprecise with very 
few bits. 
But I don't think SBR will give an improvement such that 64kbps will sound as 
good as 100kbps conventional mp3. This would mean conventional mp3 at 100kbps 
uses more than 36kbps for those frequencies that SBR extrapolates. I don't 
think so. Those parts of the spectrum can be coded with much less bitrate 
(very imprecise) without losing too much fidelity. Joint stereo codes only 
one channel for higher frequencies, because humans can't separate the 
channels at higher frequencies. I think it's more like 10% than 36%. You can 
easily verify that by analysing a few pieces of conventional mp3 for bit 
allocation. So if 100-10=90kbps is used for lower frequencies (where SBR 
doesn't do anything), how can 64kbps with SBR sound equally good? Just 
because the high frequencies sound a little better? No.

> I am also of course interested in if Ogg vorbis has plans to
> incorporate some kind of SBR-like enhancement to its format?

Probably not if it's patented and the patents will be held up.

Disclaimer: I'm not an official Vorbis project source, just a student.

--- >8 ----
List archives:  http://www.xiph.org/archives/
Ogg project homepage: http://www.xiph.org/ogg/
To unsubscribe from this list, send a message to 'vorbis-dev-request at xiph.org'
containing only the word 'unsubscribe' in the body.  No subject is needed.
Unsubscribe messages sent to the list will be ignored/filtered.