Binaural Measurement

[Hugo Zanqueta]

I am building my binaural head, probably using 2 Minidsp Umik-1. I want to know if someone have experience using binaural heads to measure room acoustic and if possible use 2 microphones in the REW.

Have a nice day

 

[Sonnie Parker]

Several of us have poked around trying it, but never been able to get it to work.

 

[Matthew J Poes]

Several of us have poked around trying it, but never been able to get it to work.

Hey Speak for yourself! I got it to work fine.

You can't do it straight into REW. What I did was output a sine sweep signal and played that back while recording using Audacity into my dummy head. This gave me a 2 channel recording. I then processed each wav recording by aligning the signals using the acoustic timing reference so that the difference between the two impulses existed after the tone, not before. I did this by measuring the difference between the two at the peak and then subtracting that from before the timing reference and adding it back in after. I then imported it to REW and I got a binaural impulse response. Well I got a left ear and right ear impulse response that has an accurate ITD. I then divided one into the other to get a difference function as well.

Any other Binaural acoustic analysis you would want to do needs to be done in a different program. Look at GNU Octave (Matlab) and some of the many MatLab programs for this.

I found some really interesting stuff in doing this, but decided none of it was something I thought anyone else would care about. It's hard to describe to folks what all this means if they don't immerse themselves in the Binaural acoustics literature.

If you have questions, let me know.

 

[Hugo Zanqueta]

Several of us have poked around trying it, but never been able to get it to work.

Thanks

Could you explain more about your experience using binaural system on the room measurement?

I have lot experience with binaural heads (Bruel & Kajaer) and binaural recording system but just using in the industry together Sound Quality and Jury Test (Psychoacoustic) for product development.

 

[Hugo Zanqueta]

Hey Speak for yourself! I got it to work fine.

You can't do it straight into REW. What I did was output a sine sweep signal and played that back while recording using Audacity into my dummy head. This gave me a 2 channel recording. I then processed each wav recording by aligning the signals using the acoustic timing reference so that the difference between the two impulses existed after the tone, not before. I did this by measuring the difference between the two at the peak and then subtracting that from before the timing reference and adding it back in after. I then imported it to REW and I got a binaural impulse response. Well I got a left ear and right ear impulse response that has an accurate ITD. I then divided one into the other to get a difference function as well.

Any other Binaural acoustic analysis you would want to do needs to be done in a different program. Look at GNU Octave (Matlab) and some of the many MatLab programs for this.

I found some really interesting stuff in doing this, but decided none of it was something I thought anyone else would care about. It's hard to describe to folks what all this means if they don't immerse themselves in the Binaural acoustics literature.

If you have questions, let me know.

Thanks

I sent a email at REW feedback email asking about that and to include other room calculations C50, C50, Bass Ratio, D50...etc

I want to to something similar you and beside study some room side unbalance (left and right). If we can do everything just in one software will be really nice.

What is your Dummy Head setup? Is it a commercial brand or DIY?

 

[Sonnie Parker]

Wayne would probably no more about what all we tried and could explain it better.

Hey Speak for yourself! I got it to work fine.

Hmmm... doesn't sound like you did. Sounds like you got something to work using two mics... just not what we are trying to do in a very simple way.

We've never been able to get two mics to work directly with REW, which is what I think he was talking about.

 

[Matthew J Poes]

Wayne would probably no more about what all we tried and could explain it better.


Hmmm... doesn't sound like you did. Sounds like you got something to work using two mics... just not what we are trying to do in a very simple way.

We've never been able to get two mics to work directly with REW, which is what I think he was talking about.

Binaural is two mics. It’s only possible with two mics.

REW is a single channel software so it can’t directly capture a binaural measurement.

I got binaural acoustic measurements to work and I used REW to process the impulse responses, just not directly in REW. I had to record it external and import to REW.

 

[Matthew J Poes]

Thanks

I sent a email at REW feedback email asking about that and to include other room calculations C50, C50, Bass Ratio, D50...etc

I want to to something similar you and beside study some room side unbalance (left and right). If we can do everything just in one software will be really nice.

What is your Dummy Head setup? Is it a commercial brand or DIY?

The Mini DSP headphone measurement rig. It’s adequate for this purpose.

 

[Matthew J Poes]

Thanks

I sent a email at REW feedback email asking about that and to include other room calculations C50, C50, Bass Ratio, D50...etc

I want to to something similar you and beside study some room side unbalance (left and right). If we can do everything just in one software will be really nice.

What is your Dummy Head setup? Is it a commercial brand or DIY?

I’ve talked to John about this as well. He said making REW a 2-Channel FFT was on his list but not a high priority. Not sure where it stands now.

You can also use it to identify the direction of reflections! I haven’t done a lot with this, but it seems feasible.

 

[luegotelodigo]

Why don't you measure both ears sequentially?

Dummy heads in room acoustics are only used to measure IACC but I think they should be used for a lot more things. For example if you equalize the effects of a reflection at a given angle you are correcting it for your omni mic that has a very different directivity than your head not to mention psychoacoustics.

 

[Hugo Zanqueta]

I’ve talked to John about this as well. He said making REW a 2-Channel FFT was on his list but not a high priority. Not sure where it stands now.

You can also use it to identify the direction of reflections! I haven’t done a lot with this, but it seems feasible.

Thank you.

I saw the MiniDST EARS. Looks really nice and affordable solution. I will email MiniDSP to get more information about EARS.
I spoke John too... I need ask him if the 2 mic solution in the REW will accept the ERAS connection (2 microphones in one USB) or will need one USB per channel.

 

[Hugo Zanqueta]

Why don't you measure both ears sequentially?

Dummy heads in room acoustics are only used to measure IACC but I think they should be used for a lot more things. For example if you equalize the effects of a reflection at a given angle you are correcting it for your omni mic that has a very different directivity than your head not to mention psychoacoustics.

That one is the simple solution, but you don't get the the time difference between left and right and more work. For 3 or 4 measurement points is not a big deal, but if you want measure 20 points will take some time.

 

[luegotelodigo]

MiniDSP ears are just a pair of ears, not a head and torso simulator.

For 3 or 4 measurement points is not a big deal, but if you want measure 20 points will take some time.

It takes a few seconds, analysing 20 measurements surely will take longer.

 

[Matthew J Poes]

MiniDSP ears are just a pair of ears, not a head and torso simulator.

It takes a few seconds, analysing 20 measurements surely will take longer.

There is actually a bigger issue and it's why I didn't do it that way in the end.

I was calculating Spaciousness measurements while talking with David Greissinger about his ideas and sharing results. We had to make sure my measurement rig was accurately capturing this and so the first thing I did was setup a test rig to calculate ITD relative to the measured distance between the mics. Remember that the MiniDSP EARS and the mics mentioned earlier are USB and have no ability to create a timing reference other than acoustically. Placing a speaker at a specified angle to each ear (creating a triangle about the head) I calculated the expected time difference. I didn't get it, it was MUCH MUCh longer. I talked with John about it and he suggested that the way that REW was using the acoustic reference wasn't designed for this. I then came up with the idea mentioned above, I manually processed the recording itself such that the acoustic timing reference was the same in each "ear" but the impulse itself (the Sine Sweep) was offset by the amount of the ITD. Doing this worked, it matched the theoretical amount of ITD the test rig suggested. Basically, using USB mics is less than ideal and adds a lot of extra work. I'm sure there is a way to use two USB mics sequentially to do this, but my approach was the simplest I could come up with.

 

[Matthew J Poes]

Thank you.

I saw the MiniDST EARS. Looks really nice and affordable solution. I will email MiniDSP to get more information about EARS.
I spoke John too... I need ask him if the 2 mic solution in the REW will accept the ERAS connection (2 microphones in one USB) or will need one USB per channel.

REW is a single channel FFT, it can't record two signals at the same time. You have to choose either the left or right ear. That is why I used a separate piece of software to capture the measurements. I did it in stereo, basically. I then imported the left and right wav files separately and used the same single sine sweep to convolve each.

If you want to preserve the timing difference accurately, you must preprocess the wav files as I mentioned in my past posts, REW will not get this right on it's own. You also can't time align them because it will adjust them to the start of the impulse, removing the delay (though that will be preserved).

You can't calculate any of the fun binaural stuff in REW, it wasn't designed for that, but you can at least calculate ITD and look at the differences in what the ears see. As noted earlier, because the head rig I use has directional response similar to human hearing, it also gives a better sense of what the ears actually detect. If you take measurements like this with two ears and keep moving the head around to each seat, you could get a sense of the bass response and deviance. This does a better job of capturing the direct vs reflected sound as would be needed for EQ.

If you want to use it to EQ much above 500-800hz, I would use something better than these mics. The HF response isn't accurate and doesn't appear correctable. There are diffraction effects in and around the ears that don't match a real human head. The Ear canal is not accurate, etc. The only way to fix that is with a real dummy head as from B&K. I can't afford those, I've looked. Even ancient old ones goes for a lot.

 

[Hugo Zanqueta]

REW is a single channel FFT, it can't record two signals at the same time. You have to choose either the left or right ear. That is why I used a separate piece of software to capture the measurements. I did it in stereo, basically. I then imported the left and right wav files separately and used the same single sine sweep to convolve each.

If you want to preserve the timing difference accurately, you must preprocess the wav files as I mentioned in my past posts, REW will not get this right on it's own. You also can't time align them because it will adjust them to the start of the impulse, removing the delay (though that will be preserved).

You can't calculate any of the fun binaural stuff in REW, it wasn't designed for that, but you can at least calculate ITD and look at the differences in what the ears see. As noted earlier, because the head rig I use has directional response similar to human hearing, it also gives a better sense of what the ears actually detect. If you take measurements like this with two ears and keep moving the head around to each seat, you could get a sense of the bass response and deviance. This does a better job of capturing the direct vs reflected sound as would be needed for EQ.

If you want to use it to EQ much above 500-800hz, I would use something better than these mics. The HF response isn't accurate and doesn't appear correctable. There are diffraction effects in and around the ears that don't match a real human head. The Ear canal is not accurate, etc. The only way to fix that is with a real dummy head as from B&K. I can't afford those, I've looked. Even ancient old ones goes for a lot.

Hi Matthew

I understood the REW limitations about binaural stuff (psycoacoustic) and thank you to share your experience. I liked your way placing a speaker in angle.
About the B&K dummy head, it is really nice equipment, but expensive with you not work with sound quality stuff in the industry. It have some tricks about the era canal and head and torso simulation.

 

[luegotelodigo]

So USB mics create a non deterministic delay in IR?

 

[Matthew J Poes]

So USB mics create a non deterministic delay in IR?

when you put it like that, it makes me think that in fact you could calculate this using the same test rig I came up with above. Since that would tell you the correct delay give the distance of the two mics, any remaining difference should be the system delay that you would normally account for in a loopback.

There is no way to input that into REW, but given how we are using it here, that isn't important anyway. So I suppose that means it is determinable but not in a simple way.

There is no way to use a Loopback with these particular USB mics, if that makes things clearer.

 

[luegotelodigo]

You don't need a loopback if the delay in both channels is always the same (acoustic delay plus a constant). I'm calculating LF in ARTA that way and works nice.

It's possible that the constant isn't constant and that would be a problem

 

[John Mulcahy]

The difficulty with USB mics is the inability to have a loopback connection that creates a timing reference. The acoustic timing reference is the solution to that for single channel measurements, but it is detected on the input you are measuring so the reference is always relative the the measured input - for a binaural measurement the reference should be one ear when measuring the other, which the current arrangement can't do. When I add stereo capture to REW it will address that.

 

[luegotelodigo]

If you're working on 2 channel option, 4 would be even better (and surely not the double of work) and would enable first order ambisonic arrays

 

[Hugo Zanqueta]

The difficulty with USB mics is the inability to have a loopback connection that creates a timing reference. The acoustic timing reference is the solution to that for single channel measurements, but it is detected on the input you are measuring so the reference is always relative the the measured input - for a binaural measurement the reference should be one ear when measuring the other, which the current arrangement can't do. When I add stereo capture to REW it will address that.

That is clear now.
But if you create a buffer array with some previous values before start the measurement you can get how many samples delay have between each side (left and right)...and maybe, I said maybe, get the same for the excitation channel to get the delay between the speaker and the first mic. after that if considering all mics and the excitation channel sample rate of 48kHz and all there synchronized you can get something close the timing reference. Maybe creating a pre-trigger and trigger to get better alignment between the array.

Sorry if I was not clear in my idea. It is just one numerical way to try synchronize and get a timing reference, I know it has some time misalignment due the computational leg the buffer data size of each channel, but considering the time between each sample rate of 2.083E-5s (48000 samples per sec) think is enough for all binaural room calculation mainly considering we are not using IEC class I data acquisition and mics.

 

[John Mulcahy]

The delays are not constant, every time a measurement is made there can be (and usually is) a different buffering delay through the OS audio chain.

 

[Hugo Zanqueta]

The delays are not constant, every time a measurement is made there can be (and usually is) a different buffering delay through the OS audio chain.

You are right... I was think about that after send the post. It make hard figure out, because the simple way is keep everything in just one hardware (inputs and outputs)... you can have a circuit (oscillator) to synchronize everything. Using 2 USB mics and PC audio card you have 2 different AD converter and one DA converter... no simple synchronization. One possibility is use just a external impulse excitation (starter gun or something the Larson-Davis BAS006 Impulsive Sound Source) to synchronize just the mics, no problem with the output channel.

Another point. Considering just the delay between both microphone, we have one theoretical delay due the distance (D in meters) between mics and the sound source position. It can change from 0 seconds when source is in front of head to (D [m]/343 [m/s]) sec when the source is placed aligned on the side of the head). Considering 0.18m the mic distance (average) the maximum source directional delay will be 5.25E-4 sec (around 25 - 26 sample) and for one specific frequency, any delay higher than that is due the room reflection / resonance. I am putting this point just to show my interest in the binaural room measurements, identify if one room have reflection and/or resonance unbalance (like one square room where the right side is anechoic with really high absorption and the left side have a really nice reflective walls)

 

[Hugo Zanqueta]

The delays are not constant, every time a measurement is made there can be (and usually is) a different buffering delay through the OS audio chain.

In my previous comments I forgot to include the phase. It helps analyse the reflection too.