[Vorbis-dev] 5.1 surround channel coupling

Richard Lee ricardo at justnet.com.au
Thu Mar 1 01:23:17 PST 2007


> I do understand that ambisonic encoding is more efficient, but there
can definitely be more information in a 5.1 mix than in a 4 channel 
ambisonic mix. In particular, I don't see how you can localize audio 
across the (film) screen as accurately without including some l=2 modes.

5.1 is better than Ambi 4.0 (which is only 3 channels) ONLY for the 5 speaker directions.  It is poor everywhere else, even for positions near the speakers.

If you want to hear 5 speakers, use 5.1

If you want to hear good sound from all directions, try Ambisonics even in its most primitive 4.0 guise.

If you want to use more channels, 5 channels will give you full horizontal 2nd Order Ambisonics and 7 channels, 3rd Order.

The beauty of higher order ambisonics is that it doesn't just make the speaker positions, but ALL directions better.

But some of us may be more interested in using plain 4 channel 1st order to get height.  Anyone know Mr Spielberg or Mr. Lucas?

>See, here's where you start sounding like an audiophile instead of an engineer. :)

I'm an engineer.  Auditory Localisation, approached correctly, is an engineering problem.

The Mk1 Human Head, is a flattened ovoid with 2 sensors slightly low and back surrounded by irregular flaps.  Inside is some processing which can take advantage of rotation & movement.

You ask yourself what CAN this do to determine directions of sound.  And compare this with research which goes back to Lord Rayleigh.  Do a few experiments yourself.

It turns out that the Mk1HH can and probably does, use a number of mechanisms.

Ambisonics attempts to get as many of these mechanisms right and where this is not possible, the "wrong" answer should be natural. ie could possibly have been from a natural source.  This makes for "good sound".

One reason why quad (and 5.1) don't work is they rely on one or two of these mechanisms but mess up the others.  So although you might get some strong localisation cues, the effect is fatiguing and unnatural cos other cues are contradictory.

OK.  I'd better come clean.  The reason why reducing 5.1 to Ambi 4.0 "sounds better" than the original is probably to do with Speaker Emphasis.  This is a BIG fault of naive 5.1 systems.  Hearing sound coming from particular speakers spoils the illusion.

A good example is the opening sequence from "The Lion King" on a good THX system.  IMHO, this has some of the best musical 5.1 sound ever.  You get good envelopment and natural sound ..... UNTIL she starts singing.  Then a noisy box (speaker) suddenly appears in front of you and you're back in your living room.

5.1 to Ambi 4.0 'blurs' the speaker positions slightly to give better results where there are no speakers so the effect is more seamless.

You probably don't want to do 5.1 into 5 channel 2nd order Ambi cos that will only give a more accurate impression of 5 boxes in the African veldt.

So an immediate saving which _might_ actually improve the illusion is to code 5.1 into WXY and 7.1 into WXYRS

Of course 5.1 into 4.0 won't work in EVERY case eg if it was important to have precise CF dialogue coming from a front box ALL the time.

But I think it would improve most 5.1 films especially if the producers have tried for good sound.

And if you encoded the sound directly into 5 channel 2nd order or 7 channel 3rd order Ambisonics WITHOUT an intervening 5.1 or 7.1 stage, you would get better, more accurate sound EVERYWHERE.

excuse the evangelising.

> Question is does the ambisonic encoding create a more effective way of exploiting the redundancy.

Yes

> If the channels are efficiently coupled the number of channels  will not determine the size ... If ambisonics results in a more compressible representation unanswered question. 

ANSWER	Yes

You can think of B-format encoding like this.  (simplified explanation mostly true)

The relative values of XYZ give you the direction cosines of the source.  W is the amplitude.  Some distance info is in the phase cos this is what happens in real life too.  For a single distant source, WXYZ are just scaled versions of each other.  This obeys superposition so WXYZ accurately encodes all zillion spherical directions.

We know Ambisonics is VERY compressible with formats which use sophisticated "lossless coupling".

The important ones tested are Dolby Digital (from Eric Benjamin, one of its inventors) and Meridian Lossless Packing (the lossless compression in DVD-A part invented by Peter Craven of the original Ambisonic team)



-- 
No virus found in this outgoing message.
Checked by AVG Free Edition.
Version: 7.5.441 / Virus Database: 268.18.4/705 - Release Date: 27/02/07 15:24



More information about the Vorbis-dev mailing list