Hi Sampo Yes your philosophical meanderings are indeed some of my concerns about image - I am not convinced that in rendering a processed B-Format file that it would decode well in all output formats - Binaural, Stereo, 5.1 etc and I am not a terribly technical DSP person so running experiments to accurately check the phase is beyond me. AS you mention, the W is a separate issue and I have thought as Joseph argues of taking a snapshot of the background noise at a quiet point (although all the recordings are quiet) for each capsule and then applying them in A-Format (tracks separated) before the B-Format conversion. In this case I am doing that only for the SPS200 for which I trust Soundfield provided me with the right decoding in their software, although I do often feel it is a few degrees off to the right, but thats another story.
I guess now that others have suggested it works for them I will apply RX to the A-Format tracks and see what happens. It does seem strange that there is not a commercially available system (that I can afford of course - i.e.. not a System 6000 solution) that does this automatically and guarantees the phase. Cheers, Garth On Aug 6, 2014, at 3:12 PM, Sampo Syreeni <de...@iki.fi> wrote: > On 2014-08-06, Joseph Anderson wrote: > >> I take the noise profile from each individual A-format channel... > > At the risk of sounding trite, what is noise? I'd argue that it isn't one > thing, and that it's pretty difficult to define with mathematical precision. > If you're talking about environmental background, then approaches like gating > A-format or some other suitable directional representation of sound is a good > idea. > > If you're talking about tape noise instead, that isn't directional at all, at > least until you get into directional masking calculations over what you can > throw away without getting caught. In that case you'd want to operationalise > what you consider noise, then find out an optimal way of extending that idea > to B-format, and do the kind of joint processing Eero suggests. > > The easiest way probably is to go with just W in the sidechain and equal > gating for all the channels in the main one. The next step would be to do the > same per frequency, and so on. However, in the ambisonic world, you'll then > bump into a third source: the mic. Since the Soundfield works on differencing > principles, W has a totally different noise profile from XYZ, and typically > it only gets worse from there as the order goes up. (Or it doesn't; that > depends wholly on the mic geometry.) > > The point is, I don't think there is a monolithic thing called "noise" which > can be just blindly "reduced". There never was even in monophonic recordings, > and the free degrees of freedom in your signal chain just multiply when you > go through stereo to ambisonic. So, you need to be careful about which > source(s) of unwanted hiss, distortion or bogus sources you're talking about, > you'll have to develop computationally tractable models of both your utility > signal and the noise, and only then can you really start to combine all of > the machinery into something which actually works/sounds good. > > E.g. when you expand/limit A-format, implicitly your noise model is a hiss > which is directional to first order and your model of the utility signal is > something like a strong, wideband directional signal near it, which makes > directional sine-to-noise masking statistics relevant. Break those conditions > and bad things will most likely happen. > > So, try your approach on a two sine test signal, separated in frequency more > than a critical band's worth. Pan one of the sines due front, and revolve the > other one around at about 1Hz and say -6dB. Then add pink noise at about > -10dB to each of the B-format channels independently. I'm rather sure that > while your approach will work beautifully for the front signal alone when > adjusted right, it'll lead to nasty, anisotropic noise pumping with the > dynamic signal in place. > > (Oh, and by the way, which A-format? As long as you're dealing with a perfect > mic and linear, time-invariant filtering operation, you don't have to think > about that because you can go willy nilly between A and B. But once you start > applying this kind of processing, every possible orientation of the mic gives > rise to a separate A-format. Which one should it be? The above example > presumes one of the capsules is facing towards the reference. It gets much > worse if you place the source directly between three adjacent capsules, in > angle space.) > -- > Sampo Syreeni, aka decoy - de...@iki.fi, http://decoy.iki.fi/front > +358-40-3255353, 025E D175 ABE5 027C 9494 EEB0 E090 8BA9 0509 85C2 > _______________________________________________ > Sursound mailing list > Sursound@music.vt.edu > https://mail.music.vt.edu/mailman/listinfo/sursound - unsubscribe here, edit > account or options, view archives and so on. _______________________________________________ Sursound mailing list Sursound@music.vt.edu https://mail.music.vt.edu/mailman/listinfo/sursound - unsubscribe here, edit account or options, view archives and so on.
