898289 - Improve interpolation of HRIRs at higher elevations

Status: NEW

(Core :: Web Audio, defect, P4)

Description

[Karl Tomlinson (back Dec 8 :karlt)]

Listen responses were measured at fewer azimuths for elevations above 45
degrees.  This would be because measurement points at the same number of
azimuths would become closer together at higher elevations.  To simplify the
HRTF implementation all HRTFElevations have the same number of azimuths.

If elevation is > 45, then missing azimuths are picked up from lower
elevations rather than interpolating.

This causes a discontinuity in effect when the source point passes from left
to right overhead, observable near the centre in
http://people.mozilla.org/~karlt/panner-horizontal.html

http://people.mozilla.org/~karlt/panner-vertical.html does not have this
effect when the source passes from left to right in front of the listener.

Interpolating from neighbouring azimuths in the same elevation will likely
give better results, especially at elevation = +90 degrees where there is a single HRIR.

(The equalpower panning model has a similar jump, which is due to using axial
 symmetry around a vertical axis instead of around a horizontal axis through
 left and right ears.  The vertical axis produces a discontinuity along the
 entire axis, while the horizontal axis would have a discontinuity only at the
 listener.  A object passing from left to right overhead is expected to sound
 similar to an object passing from left to right in front or behind, in a
 simple model.)

Comment 1

[(no longer active)]

Is this a bug in the Blink imported code?

Comment 2

[Karl Tomlinson (back Dec 8 :karlt)]

Yes, this is an issue in the Blink imported code.
The patches in bug 865241 retain that behaviour.

Comment 3

[(no longer active)]

It would be great if we can upstream our fix here then!

Comment 4

[Bulk Bug Changes for mreavy's org]

Mass change P3->P4 to align with new Mozilla triage process.