Recording Impulse Responses for Speaker Virtualization

[sander99]

Could you elaborate on what did you do exactly? What I gathered from this is that your measured a mono system and somehow that creates a sound stage. I'm sure this is not the case. How many speakers did you measure? At which positions? Did you do something extra to the output HRIR file that Impulcifer doesn't do? Do you need some kind of special DSP processing to make this work with Impulcifer created HRIR? Did you have some kind of special DSP processing for the speakers during the measurement? I'm baffled.

If I may take a guess: he probably did what boiles down to the same as what @Erik Garci did with his A8: Erik measured a 2 channel PRIR using one center speaker. He sent the left and right channel sweeps to that one speaker. When the A8 measured the left channel he muted the right in-ear microphone. When the A8 measured the right channel he muted the left in-ear microphone.
So what he gets at playback of a stereo source is this: the left ear "hears" the left channel of the input source being played over the center speaker, and the right ear "hears" the right channel of the input source being played over the center speaker. It figures that for the mono-component in the signal it sounds very natural, because for that mono-component the total works the same as if that mono component had been sent to a "normal" virtual center speaker. Non-mono parts of course will have ILD and/or ITD clues. The further out of center sounds are placed the further there will be a mismatch between the actual HRTF filtering (based on the position of the center) and the HRTF filtering that would correspond with the placement in the "soundstage".

Note: I never tried any of this myself, and I do realise it is higly subjective and not at all a-priori logical to apply this to recordings that are made with a different usage in mind.

 

[dwk]

Yes, @sander99 is basically correct - just use Impulcifier to perform a normal binaural measurement of the FC channel. Then run the analysis as normal, and Impulcifier will generate a hrir.wav with only FC-left and FC-right elements being non-zero. Pull these 2 channels out into a normal stereo .wav, and load into the JRiver convolver so that FC-left is applied to the left channel, and FC-right is applied to the right channel. (I actually tweaked the impulcifier code to spit out the 2 filters into ambio.wav directly, but you can use audacity). JRiver makes this easy since it supports a stereo wav file as a convolution filter directly - you don't have to generate a config file for it.

And yes - what this does is trade inaccurate HRIR for mono/center sounds which you get in normal stereo for HRIR inaccuracies in L/R separated content. This seems to be a reasonable trade with center-dominated content, but maybe not for more general use.

So, in terms of an experiment this 'works' to a degree, but doesn't actually reproduce the subjective impressions of a true ambio system since the L/R sounds collapse in a way that they don't over speakers. Not entirely sure why - I may continue to do some experiments. It's still the best headphone sound I've had, at least for the voice+guitar stuff that makes up the bread and butter of my listening. I'm going to compare to the conventional stereo measurement I also made - it's entirely possible that the stereo version will be so good that I'll abandon the 'ambio simulation' route, but we'll see. Either way, the fact that $350 in headphone gear can sound THIS good is remarkable.

The other reason I was interested in trying this is that it's a basic 2-channel process,and could be used with something like the minidsp HA-DSP which doesn't really have enough horsepower to do a proper 2x2 convolution, or maybe even on an iPhone so I could have a system for use at work. Not entirely sure whether this will come to pass, though.

One other interesting thing I'll have to look into - when I measured my 'big system' I did not experience the same lack of bass as I did with the R3s measured on my desktop. Subjectively I don't find the R3s lacking bass when played normally (although the main system is defnitely better), but I had to boost the bass quite a bit to get them sounding OK. No such problem with my main speakers - great bass captured right in the measurement.

 

[musicreo]

Yes, @sander99 is basically correct - just use Impulcifier to perform a normal binaural measurement of the FC channel. Then run the analysis as normal, and Impulcifier will generate a hrir.wav with only FC-left and FC-right elements being non-zero. Pull these 2 channels out into a normal stereo .wav, and load into the JRiver convolver so that FC-left is applied to the left channel, and FC-right is applied to the right channel.

I 'don't understand this concept. I will hear left only on the left headphone speaker and rigth only on the right headphone speaker? Is this not the opposite of what I want to get from impulcifer?

 

[sander99]

I 'don't understand this concept. I will hear left only on the left headphone speaker and rigth only on the right headphone speaker? Is this not the opposite of what I want to get from impulcifer?

If your goal is to realistically binaurally simulate loudspeakers in a room then you are right. In this concept impulcifer (or a realiser) is used for something completely different, an alternative way of listening to stereo recordings (with HRTF filtering and room influence but without cross talk), different from normal (or realistically binaurally simulated normal) loudspeakers (HRTF filtering and room influence & cross talk), and different from normal (unprocessed) headphone listening (no HRTF filtering or room influence & no cross talk). Whether or not this is usefull or "an improvement" is debatable and subjective, lets just say that some people like it with some recordings.

 

[dwk]

Once again, agree with @sander99. What I'm doing is an unconventional/alternate use of the Impulcifier capabilities to experiment with some different ideas. Read up on Ambiophonics and the discussion (perhaps 'cult' isn't too far from the truth) to better understand why I was interested in trying it.

Having had a day or so to compare the 'ambio' single-center-channel measurement to the 'normal' stereo measurement, I think I have to conclude that the conventional stereo measurement is generally better. The 'ambio' is preferable on certain minimalist vocal recordings, but as soon as you add much instrumentation the stereo version becomes much more credible. I'm still not entirely sure why the ambio idea wasn't more successful, and may continue some experiments - maybe getting a bit of separation and trying a 20-degree 2-channel dipole for example. I think the conventional stereo version will be my normal setup for now though.

 

[Dogway]

Found this interesting to share: Road to PS5

 

[jaakkopasanen]

Found this interesting to share: Road to PS5

Cool! They certainly seem serious with 3D audio. Of course it will be years before they have decent HRTF synthesis working if they get it working well at all. But anycase I'm happy to see big companies starting to tackle this problem.

 

[Dogway]

Yes, things like this going mainstream is very good to raise awareness in audio tech. People just don't care or know enough. Very surprised to see him talk about HRTF but I'm not sure how personalized HRTFs could work with a few photos (like Super X-Fi).
I have been tempted for some time to sculpt my ear in 3D and see if there's something that could be done programmatically with that, found E.A.R addon for Blender. It's something to research on.

 

[johnn29]

Thanks for posting the Sony link - fascinating. It's nice that it's gone from military simulation environments to more main stream gaming now. Sony's approach is novel - I like that they'll have some sort of test to see which of the 5 matches you well and are eventually looking to develop a custom version.

They've recently done the 360 audio with ear mapping - hope this makes it to that. It's crazy that we're on the verge of simulating the perfect loud speaker in the perfect room that'll blow away any real setup. Great time to be an audiohead.

 

[johnn29]

jaakko I'm planning on cutting my ear hooks and gluing ear plugs to my mics like you did. Is this still something you advice or do? Now that we're all at home and social distancing I might as well try and nail some more HRIR's

I'm hoping this method leads to much easier recordings that are repeatable - which would imply they are more realistic. I'd like to stop relying on transform for my over-ears and rely on headphone compensation. But historically it's actually been hard to nail a good recording due to channel balance issues and the damn mics that keep falling out or changing position.

 

[jaakkopasanen]

jaakko I'm planning on cutting my ear hooks and gluing ear plugs to my mics like you did. Is this still something you advice or do? Now that we're all at home and social distancing I might as well try and nail some more HRIR's

I'm hoping this method leads to much easier recordings that are repeatable - which would imply they are more realistic. I'd like to stop relying on transform for my over-ears and rely on headphone compensation. But historically it's actually been hard to nail a good recording due to channel balance issues and the damn mics that keep falling out or changing position.

Using ear plugs instead of the hooks makes life so much easier. The mics stay in place without any worry of moving or falling out. I also find it easier to place the mics in the same location time after another. I think I would trust it right now to make measurements and then measure another pair of headphone later even though the mics were taken off in between. So yes, I would recommend cutting the hooks.

 

[lowdown]

Using ear plugs instead of the hooks makes life so much easier. The mics stay in place without any worry of moving or falling out. I also find it easier to place the mics in the same location time after another. I think I would trust it right now to make measurements and then measure another pair of headphone later even though the mics were taken off in between. So yes, I would recommend cutting the hooks.

Had the same experience. Trying to use the mics with hooks was very frustrating with the precarious, unstable positioning. I was also a bit concerned the hooks could alter my pinnae's frequency reflections, though I have no evidence, just a thought. I cut about 1/3 off of some foam earplugs to shorten them, glued on the back of the mics and it made all the difference. Final results with Impulcifer are incredible.

 

[johnn29]

Thanks guys, going to do that on Monday.

The idea of doing only a single speaker recording and multiple headphone compensations later in time is going to be awesome.

 

[johnn29]

So I cut the hooks and glued onto foam plugs. SO MUCH EASIER! And native channel balance is perfect, no processing needed.

Edit: Tried measuring headphones after the initial recording - results are great!

 

[pfzar]

The first comes from a long-standing feeling that the headphone guys and ambiophonic guys need to talk to each other. Ambiophonic guys are spending a LOT of time and money to set up a speaker system that eliminates crosstalk whereas the headphone guys seem to be constantly looking for new ways to introduce crosstalk. This led me to a theory that what the headphone guys really want is HRTF and not crosstalk.

This is an incorrect assumption of what ambiophonics is. By having the speakers closer together you are shortening the path of the contralateral and the ipsilateral. You then run the signal through a feedback delay network. In this manner, you are creating a cross-talk cancelation network.


Happy to see so much excitement about HRTF and spatial audio for headphones. Psychoacoustics, and HRTF's are my primary research responsibilities. Fun times indeed with all the new things happening here.