Recording Impulse Responses for Speaker Virtualization

[Joe Bloggs]

It's where the channels bleed into each other. For example a centre channel tone will also be heard from the back left channel, perhaps with a slight delay and echo

Has this been ascertained by replacing the back left channel HRTF with silence?

In other words, do we have proof that the signal is getting fed to that channel somehow or does it only *sound* like it's coming from that direction?

 

[musicreo]

In other words, do we have proof that the signal is getting fed to that channel somehow or does it only *sound* like it's coming from that direction?

The channel routing is done by EQ-APO/ Hesuvi so it should not possible that channels are fed the wrong way.

 

[johnn29]

It only sounds like that. It's not actually channel crosstalk. I said it was bass related because it only occurred for me in one measurement I took without a sub and used the virtual room correction to boost bass massively. But I believe it's not to do with that exactly

 

[jaakkopasanen]

Here is a demonstration of what I'm experiencing when I talk about this problem: https://drive.google.com/open?id=1d91VZI1cyfbXNBBjYIwwWPdc-4S-2_Di

You need to use headphones when listening to the 8_Channel_ID_binaural.wav. The low frequency ringing can be heard best with center channel on the right side. I added plots for the center channel there as well but I'm not seeing big difference between left and right sides that would explain what is going on. The plots are of the output impulse responses with all the processing already done.

 

[Joe Bloggs]

Here is a demonstration of what I'm experiencing when I talk about this problem: https://drive.google.com/open?id=1d91VZI1cyfbXNBBjYIwwWPdc-4S-2_Di

You need to use headphones when listening to the 8_Channel_ID_binaural.wav. The low frequency ringing can be heard best with center channel on the right side. I added plots for the center channel there as well but I'm not seeing big difference between left and right sides that would explain what is going on. The plots are of the output impulse responses with all the processing already done.

I guess you guys are boosting low frequencies too much with speakers that aren't reproducing them in the sweep. The bass would then be of the right magnitude but random phase that's all over the place. You'd do better to ignore the recorded signal for frequencies below a certain cutoff magnitude and synthesize the frequencies in the output impulses to be whatever magnitude you want and phase that matches the rest of the impulse.

 

[jaakkopasanen]

I guess you guys are boosting low frequencies too much with speakers that aren't reproducing them in the sweep. The bass would then be of the right magnitude but random phase that's all over the place. You'd do better to ignore the recorded signal for frequencies below a certain cutoff magnitude and synthesize the frequencies in the output impulses to be whatever magnitude you want and phase that matches the rest of the impulse.

Thanks for the insights! Sounds like the alternative way to do this would be to create a mixed phase FIR filter for the room correction which corrects both the magnitude and the phase. Not sure which is better or easier.

 

[Joe Bloggs]

Thanks for the insights! Sounds like the alternative way to do this would be to create a mixed phase FIR filter for the room correction which corrects both the magnitude and the phase. Not sure which is better or easier.

I would think that the phase of frequencies that are measured correctly (with actual signal from the loudspeaker) contributes to the room feel of the impulse while phase of rubbish frequencies (that the speakers can't put out and where the mic is just picking up background noise) would best be ignored. So unless you can build a mixed phase FIR filter that takes this into account, I would say the other approach should be better / easier.

It may also be best to have an option for the user to manually enter the range of frequencies to be considered, as bass background noise can sometimes be strong enough that simply considering the volume picked up by the mic wouldn't be appropriate for checking what frequencies were actually played.

 

[jaakkopasanen]

It's not just the room correction eq boosting bass that's causing this because the ringing / bleeding / whatever is present also in HRIRs that don't have any EQ built in. See here for newer results: https://drive.google.com/open?id=1BCcp7_hJ9WFn5UNKEsAM9UFymvmZtxeu. There are versions of the same file with and without room correction, headphone compensation and channel balance correction. All of the files have some degree of low frequency ringing.

 

[Joe Bloggs]

It's not just the room correction eq boosting bass that's causing this because the ringing / bleeding / whatever is present also in HRIRs that don't have any EQ built in. See here for newer results: https://drive.google.com/open?id=1BCcp7_hJ9WFn5UNKEsAM9UFymvmZtxeu. There are versions of the same file with and without room correction, headphone compensation and channel balance correction. All of the files have some degree of low frequency ringing.

The low frequencies that aren't reproduced by the test speakers will be problematic regardless of these, if the mic picked up background noise at these frequencies. What frequencies does it start to ring and what frequencies do you reckon your speaker can go down to?

 

[jaakkopasanen]

The low frequencies that aren't reproduced by the test speakers will be problematic regardless of these, if the mic picked up background noise at these frequencies. What frequencies does it start to ring and what frequencies do you reckon your speaker can go down to?

I haven't found a good synthetic test signal for this so I don't know the frequencies where it rings. However the channel ID has male voice which should not be less than 80 Hz and this particular one is probably around 100 Hz or more. My speakers (Dynaudio Focus 110) start to roll off at 70 Hz which is lower than the lowest male voices so this shouldn't be the issue here.

 

[johnn29]

The virtual crosstalk/ringing doesn't happen to me when I'm recording my system with a sub, mine's flat to 6hz. It only happens when I don't use a sub, so it has to be something about producing those frequencies with your speakers right? Can we do a "direct bass" feature like the Smyth has where bass is just passed through?

 

[jaakkopasanen]

Last bits of room correction have now been implemented. I added generic room measurement which will be used for speaker-ear pairs that don't have the specific named measurements. room.wav is the file where these measurements should go and recorder.py can now add tracks to existing WAV file with parameter --append so it's possible to run recorder.py for the same room.wav file multiple times, each time moving the measurement microphone a bit.

I added two methods for combining the frequency responses of these multiple room measurements. The first is a simple averaging method and the second uses the absolute minimum difference and only if all the measurements are on the same side of the 0 dB level. So with the conservative method the error will be 0 dB at 1 kHz if the errors of the 5 different measurements at 1 kHz are -0.4 dB, 0.3 dB, 1.9 dB, 1.1 dB and 3.0 dB even though the average is clearly above zero. This makes it a lot safer since it ensures that the correction won't make any of the measurements worse. See example graphs here: https://imgur.com/a/jcHpaBv and the details for usage can be read here: https://github.com/jaakkopasanen/Impulcifer#room-correction. I still need to write room measurement guide but now that the implementation is done I can actually start on it.

Another smaller update is that the hrir.wav and all responses wav files now include LFE tracks. The LFE tracks are silent but I decided to add them there to be compatible with other systems (like ffmpeg) which expect them to be there. The speaker order is now: FL, FR, FC, LFE, BL, BR, SL, SR. Every odd numbered track (1st, 3rd, ...) is for left ear and every odd numbered for right ear.

The virtual crosstalk/ringing doesn't happen to me when I'm recording my system with a sub, mine's flat to 6hz. It only happens when I don't use a sub, so it has to be something about producing those frequencies with your speakers right? Can we do a "direct bass" feature like the Smyth has where bass is just passed through?

Does the crosstalk / ringing sound the same for you? Could you perhaps share a pre-processed file where you hear this or share the measurements without sub? I would like to make sure we are experiencing the same thing.

What is this "direct bass" feature? What does bass pass through mean?

 

[sander99]

What is this "direct bass" feature? What does bass pass through mean?

With direct bass the bass below some crossover frequency is not binauralised but directly mixed into the headphone signal. But of course bass from each input channel goes to both headphone channels, and properly delayed with the latency of the binauralisation process I assume. I vaguely remember Stephen Smyth saying in a podcast that in the A16 they refined this function with adding left-right or right-left relative delays.
The result would be very dry precise bass, actually just as dry and precise as your headphone can do it.
(For people who find it too dry, or unnaturally contrasting the behaviour of the other frequencies, maybe an option to add some slight artificial reverb in the bass would be a nice feature? The Smyth don't have that as far as I know.)

In your case, as you only generate the impulses and let the rendering do by third party software like HeSuVi, you would have to manipulate the impulse responses (or the sweep responses) somehow, but I won't have to tell you that of course.

 

[johnn29]

Does the crosstalk / ringing sound the same for you? Could you perhaps share a pre-processed file where you hear this or share the measurements without sub? I would like to make sure we are experiencing the same thing.

What is this "direct bass" feature? What does bass pass through mean?

Sander summarised the direct bass feature great. It just might be useful to try out.

I'll get the files over to you tomorrow - I'm abroad right now but I have my trusty Galaxy Book with me and keep my recordings so I can update my hrir's wherever I am! I'll also listen to your ones to see if we're talking about the same thing.

 

[Joe Bloggs]

I haven't found a good synthetic test signal for this so I don't know the frequencies where it rings. However the channel ID has male voice which should not be less than 80 Hz and this particular one is probably around 100 Hz or more. My speakers (Dynaudio Focus 110) start to roll off at 70 Hz which is lower than the lowest male voices so this shouldn't be the issue here.

https://www.dropbox.com/s/4zrci9cd7a8mz9s/SineGen.exe?dl=0

Try this

It could conceivably be a strong resonance mode at that frequency in your room