Creation of an impulse response from external software with REW

[Matthew J Poes]

Hi John,

I've been chatting with you privately about this, but figure it may be worth posting publicly, without getting into all the details.

As you know, I'm trying to get an accurate impulse response from the binaural recording. I was originally using your log sine-sweep with timing reference, which I turned into an external wav file, and then recording the binaural measurement of a single speaker in room through Audacity. Importing the Left and Right channels separately into REW and trying to use L/R to create the difference file, representing ITD and ILD. It didn't work, what I got was mostly garbage, and I've been told that it's likely a problem in the FFT. The words used were "There are likely 0's in either the left or right impulse not present in the other that would cause the math to "blow up."

It also turns out the test tone I used doesn't have the properties needed to assess the room as I intended. What I was told to do instead is to use a log swept sine with reverb. Here is a quote:
"To make an IR I record a sine sweep, getting something that looks like a sine sweep, but with reverb added. I then take the sine sweep I used as a source to the loudspeaker, and reverse it in time. Audition will do this. I then convolve the measured sweep with the time-inverted source sweep to get an IR. This is magic. And there is a trick."

If I import the wav files I made using this test tone into REW as I did before, is this essentially what it does? Such that I could export the resulting impulse response and use it as needed to analyze envelopment and spaciousness? I don't have audition nor do I want to spend the money. I can't figure out how to use Audacity to reverse in time the measurement nor convolve the two together, so if REW handles this automatically in creating the IR, then I don't have to keep wasting time trying to figure this out.

I know you said this would happen someday (and not soon), but when you do bring in 2 channel (or hopefully simply multi-channel), it would be amazing if it could automate the measurement of some things like this for dummy like me. There is a lot of amazing academic work out there for measuring spaciousness, envelopment, etc. that go beyond ISO 3382 and have real relevance for small rooms, but its all buried in Mattlab code and AES papers. I'm swimming in this stuff barely able to stay afloat in trying to do this with freely available tools.

 

[John Mulcahy]

The problem isn't getting impulse responses (that is what REW already does whenever it makes a measurement), it is preserving relative timing of the binaural channels. The acoustic timing reference isn't suitable for that as it only has the desired effect for different measurements taken from the same mic location rather than two different mic locations. It could be done fairly easily with analog mics, using a loopback to provide a timing reference and capturing each mic channel separately, but to do it with USB mics (e.g. EARS) needs me to add multichannel capture support.

As an aside, REW's levels meter has a feature that shows the relative amplitude and phase of the input channels. It is intended for checking electronic signal paths with single frequency tones, might be interesting to see how that behaves. Just out of curiosity, not suggesting it as a measurement method.

 

[Matthew J Poes]

The problem isn't getting impulse responses (that is what REW already does whenever it makes a measurement), it is preserving relative timing of the binaural channels. The acoustic timing reference isn't suitable for that as it only has the desired effect for different measurements taken from the same mic location rather than two different mic locations. It could be done fairly easily with analog mics, using a loopback to provide a timing reference and capturing each mic channel separately, but to do it with USB mics (e.g. EARS) needs me to add multichannel capture support.

As an aside, REW's levels meter has a feature that shows the relative amplitude and phase of the input channels. It is intended for checking electronic signal paths with single frequency tones, might be interesting to see how that behaves. Just out of curiosity, not suggesting it as a measurement method.

I believe for a Binaural measurement to be captured accurately you would need 3 channels to use a loopback. Someone else suggested that.

Ok so back to the problem here, obviously with the method I am using to record the binaural recording, I have the ability to preserve the relative time in the original recording, but you are saying that isn't true of the impulse anymore. Now it gives you an estimate of the IR delay, would that be the IR delay relative to the first sample (if there is no acoustic reference)?

 

[John Mulcahy]

I believe for a Binaural measurement to be captured accurately you would need 3 channels to use a loopback.

Two channels is sufficient, the measurement channels don't need to be captured at the same time as the loopback provides a consistent reference for each.

it gives you an estimate of the IR delay, would that be the IR delay relative to the first sample (if there is no acoustic reference)?

No, the delay estimate is relative to a minimum phase version of the response and in producing the impulse response there would already have been shifts applied to the captured data.

 

[Matthew J Poes]

Two channels is sufficient, the measurement channels don't need to be captured at the same time as the loopback provides a consistent reference for each.

No, the delay estimate is relative to a minimum phase version of the response and in producing the impulse response there would already have been shifts applied to the captured data.

Thanks John. I need to clearly move on to other options then.

Ok one last plug for my earlier idea. What if I used the acoustic timing reference in the recording to calculate the delay to the first sample. Then Calculate the difference and plug that difference into the shorter file as silence. Then the timing reference would be in the same relative place and the difference between the impulses themselves would reflect the difference in time between ears. I know you said that initial calculation would only be the ITD at high frequencies. If used simply as a reference point like this, wouldn’t it preserve the full frequency ITD at the point of the impulse itself?

If that doesn’t work I could simply plug the relative difference into the impulse response file using Audacity. Any reason that wouldn’t work?

 

[John Mulcahy]

If you insert a gap between the timing signal and the rest of the captured data corresponding to the delay difference that should preserve the delays between the channels.