[openal] Speaker test

[Chris Robinson]

On 06/04/2017 01:14 PM, Anders Genell wrote:
> So, does OpenAL Soft use a matrix of predefined possible speaker
> positions when decoding b-format (there are som standard ones related
> to the wave-x format in some way, isn't there?) or how is the
> b-format otherwise decoded to a speaker array?

Basically it takes an accumulation of how much each B-Format channel 
contributes to the direction the speaker is in. That's a pretty gross 
oversimplification, but it's the simplest way to explain it that I can 
think of.

The general concept is, given a sound panned in B-Format to some 
direction, all the speakers contribute to make a physical plane wave 
coming from that direction. The more speakers you have, and the higher 
the ambisonic order, the more accurate the plane wave can be.

> You say a multi-channel audio file may or may not have corresponding
> output channels, but if the number or channels in the file or stream
> exactly matches the number of channels in the output format set in
> alsoft.conf should they then at least match channel to channel?

I tried doing that once, but I found there was a perceptual volume 
difference when an input channel is matched directly to an output 
channel, compared to being panned. At least for surround sound output. 
And for stereo, there's the recently-implemented stereo_angles extension 
that allows rotating the stereo channels, so even if the channel names 
match, their intended positions might not.

And things like HRTF, which is stereo output but uses filters to provide 
consistent reproduction of multi-channel sources (you wouldn't want to 
directly match input to output with HRTF anyway, unless the sound is 
already made for headphones).

One of the ideas I currently have in mind is something where you can 
query the number of output channels, get text labels for each channel, 
and some method to "link" a source to play on a specific output channel 
instead of going through the panning stuff. This should be reasonably 
flexible since it wouldn't restrict OpenAL to using a list of predefined 
formats, or a list of predefined channels. Not too keen on the idea of 
text labels to identify channels, but I don't think predefined enums 
would be the best idea for ease of extendability.

> Also, what kind of b-format decoder does OpenAL Soft use? Would it be
> possible to have it decode to a horribly skewed speaker setup that
> would need both angular and time/phase correction?

Not sure about angular correction, but time/phase correction on speakers 
is possible. Simply provide an .ambdec config file that specifies the 
proper speaker distances, and OpenAL Soft will apply the necessary 
delays and gains to make them sound equidistant.