Subject: Gratuitous Rant on HRTFs/Ambisonics, was RE: [linux-audio-dev] surround plugin port ?
From: Richard W.E. Furse (richard_AT_muse.demon.co.uk)
Date: Sun Nov 19 2000 - 03:57:28 EET
QuiXound sounds like a crosstalk/HRTF system. This isn't a field I've been
concentrating on - these approaches improve the spatial perception at a
perfect listening point (or in headphones) at the expense of the integrity
of the overall soundfield - so strange things tend to happen for off-centre
or angled listeners. Instead, I've been improving the image by using extra
speakers so the listener can wander around the soundfield - the sound is
essentially coming from the right place to start with so there's less to
fix (given various 2D/3D skew effects similar to watching TV off-axis that
the brain compensates for). For those of you who don't know Ambisonics
(probably quite a few) it is based in mathematical rigour (mostly), unlike
5.1 and suchlike. You can buy kit that mixes Ambisonics down to 5.1 but not
vice versa.
<particularly techie rant>
Having said that, mixing down an Ambisonic soundfield to a stereo HRTF
format is not only doable but easy - I've been meaning to do it for a while
- it boils down to a few channels of trivial convolution against fixed
impulses. However, to do this one needs to generate these impulses - which
is again fairly trivial as a process (integration, probably numerical, of
your chosen HRTF set over the unit sphere weighted by each spherical
harmonic with optional extra faff if you want to do crosstalk cancellation
for speaker rather than headphone listening). The *problem* is getting hold
of a suitable set of HRTFs that cover the entire unit sphere. I've played
with MIT's KEMAR set, but it colours the sound way too much, there are too
few spatial samples to get around phase ambiguities in the responses
(important during the integration) and it doesn't cover some of the lower
elevations. I'm hoping someone will come up with a better set of recordings
or a set of formulae that provide an OK general model some time soon.
</particularly techie rant>
Hmmm. That's probably more than you wanted to know. To summarise, Ambiso
nics can be used to do this stuff, but it requires a good set of HRTFs to
do the Ambisonic->stereo mixdown. Anyone have any suggestions? In fact,
using Ambisonics as your internal format and then mixing down to HRTF will
be significantly faster than using HRTF on all channels once you have
enough channels to overcome the overhead (possibly only 2 or 3 HRTFs for
first order Ambisonics, 3 or 4 for second order). The extra cost of the
Ambisonic encoding and mixing is tiny when compared to the cost of doing
the extra HRTFs, particularly if you're interpolating spatially. Plus using
Ambisonics as your spatial encoding allows mixdown to phase-safe mono,
conventional stereo or 5.1 - and your encoding can be played over a
full-blown Ambisonic array! You can even spin and generally mess around
with the soundfield while encoded.
To summarise the summary, if someone can get me a good (legal!) set of
clean HRTF recordings/functions then we can produce some very serious
tools! And Ambisonics is fantastic...
- And in the meantime, getting QuiXsound working as a LADSPA plugin would
be great. It's probably a rather more straightforward approach. I'm also
rather curious to see what they're up to. Must Try To Find Time...
--Richard
-----Original Message-----
From: Paul Barton-Davis [SMTP:pbd_AT_Op.Net]
Sent: Saturday, November 18, 2000 5:07 AM
To: linux-audio-dev_AT_ginette.musique.umontreal.ca
Subject: [linux-audio-dev] surround plugin port ?
i just saw this on freshmeat ... perhaps someone would like to port it
to LADSPA ? its for XMMS now, so it can't be *too* hard. or maybe
richards ambisonics stuff is already the right stuff for this.
does xmms have any other cool plugins that we should make available ?
could we ever convince xmms to use LADSPA ?
--p
QuiXound 3D Surround Studio is an effect plugin for XMMS which
applies several effect envelopes on the input signal and outputs
real 3D surround audio. 3D surround audio makes the source of
sound seem to be beyond the actual spatial position of the
speakers.
Changes: Everything has changed; it is now 3-in-one effect, with
an effect chainer.
This archive was generated by hypermail 2b28 : Sun Nov 19 2000 - 04:39:37 EET