Innovative personalized 3D sound solution for headphones: introduction and beta release

[myHRTF]

Hello @olieb , @Robbo99999 and @ob1 ,

First of all, thank you very much for taking quite some time to go through the personalization, trying out the different music pieces and settings, and writing all the feedback. I very much appreciate that!

Let me answer common points in this response. In addition, I will respond to all of you individually to address the remaining points.

I fully understand the need to test this with your own music. Unfortunately, the simplest way - uploading of own music files - is not possible since in theory this could be used to just copy the filters by uploading a Dirac signal. However, I have a special offer for you: send me a PM and we can arrange for some songs to be pre-filtered offline.

Two of you find that the reflections are too strong. olieb pointed that out directly and ob1 finds it a bit too diffuse/imprecise/blurred, which points in the same direction. Maybe it would be better to have both a room with dry acoustics like a control room and a room with more reflections...?
Let's wait for some more feedback on this.

Philipp

 

[myHRTF]

Hi, thanks for the answers. What rig did you measure the headphones on? How many units of each headphone did you measure? Did you measure the headphones with new pads, worn pads or something inbetween?

We use both industry standard equipment and measurement equipment that we developed for that purpose. In order to avoid that this thread changes to a discussion about measurement rigs and target curves, please kindly allow me to spare you from more details on this.

In the beginning we would measure up to five samples of the same headphone model. But our own measurements and other measurements published in the internet show that differences between samples are very small, at least for those headphone models that we support.

The pads were new or almost new, no worn pads.

 

[jaakkopasanen]

Thanks @myHRTF for creating this!

I tried the demo and I am positively surprised! That being said, I wasn't expecting much because I have heard many virtualization solutions that don't do a proper customization. With myHRTF I was able to get a decent out of the head front localization, although the localization is not very accurate. It feels more like "somewhere there in front of me", but that's already better than almost any other I've tried.

The tonality was a disappointment. I'm not very good at identifying what's wrong specifically, or describing it, but definitely the balance is not right. The demo sounded a bit "tinny", and was missing a lot of the lower midrange.

All being said, I think you're onto something with this. May not produce the same accuracy as Smyth Realizer or Impulcifer, but is also a lot less hassle. For proper commercialization and acceptance with greater population, the tonality needs to be improved.

 

[Robbo99999]

Hello @olieb , @Robbo99999 and @ob1 ,

First of all, thank you very much for taking quite some time to go through the personalization, trying out the different music pieces and settings, and writing all the feedback. I very much appreciate that!

Let me answer common points in this response. In addition, I will respond to all of you individually to address the remaining points.

I fully understand the need to test this with your own music. Unfortunately, the simplest way - uploading of own music files - is not possible since in theory this could be used to just copy the filters by uploading a Dirac signal. However, I have a special offer for you: send me a PM and we can arrange for some songs to be pre-filtered offline.

Two of you find that the reflections are too strong. olieb pointed that out directly and ob1 finds it a bit too diffuse/imprecise/blurred, which points in the same direction. Maybe it would be better to have both a room with dry acoustics like a control room and a room with more reflections...?
Let's wait for some more feedback on this.

Philipp

I'll PM you some requests for tracks that I use to test my EQ's. Thanks. EDIT: PM sent.

We use both industry standard equipment and measurement equipment that we developed for that purpose. In order to avoid that this thread changes to a discussion about measurement rigs and target curves, please kindly allow me to spare you from more details on this.

In the beginning we would measure up to five samples of the same headphone model. But our own measurements and other measurements published in the internet show that differences between samples are very small, at least for those headphone models that we support.

The pads were new or almost new, no worn pads.

Ok, that's your prerogative re being vague about what measurement rig you are using to measure the headphones, I don't like to hear that, but I don't mind. Re headphone unit to unit variation it does differ between different headphone models, some are better than others, but it's a bit tricky if you're just relying on one unit for your work, but I can totally understand the cost side of it. I think it's good that you're using new or almost new pads, that's ideal.

 

[myHRTF]

First of all, you need a stop button for the sound on your website.

Good suggestion, let me think about that.

There is quite a lot of reflections even with studio and 1m distance selected. Way too much for my taste.

By chance, are you a professional sound engineer or similar if I may ask? Reason I ask is that people with this background tend to prefer less reflections. In any case, I consider to add a room with significantly dryer acoustics.

(Switching the playback options is tedious as there are dependencies between the menu items in different menus.)

The personalization settings are not intended to be used as kind of EQ to adapt the sound, since once they are set to your anatomy correctly, other settings can only decrease the quality.

Violins (and everything else with treble) sounded very piercing.

Thank you for this feedback, it is particularly important. Let me check if this is connected to the headphone model used.

A room virtualisation should not turn the sound balance upside down

I couldn't agree more with you on this one. Having said that, the sound from good speakers in good room acoustics is different from headphone sound, hence some of the changes in timbre are intended and required.

For any kind of meaningful comparison the bypassed signal should have (optional) EQ too.

Do you mean that you would like to see an optional bypass with EQ to Harman target curve?

And generic HRTF solutions exist for quite a long time.

Let me assure you that this is not a generic HRTF solution. Generic HRTF solutions don't work properly unless your HRTF happens to be sufficiently similar to the HRTF of the dummy head. If not, they tend to cause front-back and up-down confusion that need to be resolved with head-tracking.
Otherwise we could just use the dummy head recording method and call it a day.

As I understand, there are algorithms to create an HRTF from multiple photographs of the ears/head. To me it would be much more promising to offer that as a service (in .sofa format).

Typically, the SOFA file contains the direct wave only, i.e. you hear the music as if you would sit in front of speakers inside an anechoic chamber. Hence, you need to add reflections. And this is where the real challenge begins in my experience.

 

[myHRTF]

My biggest first impressions are that the music doesn't sound wrong tonally, so I think that's roughly right, and it does to me sound like you're in a room listening rather than listening on typical headphones - to me it sounded like listening to a live performance in a room more than recorded music.

Thank you for this positive feedback. If you mean by live performance that you perceived the virtual instruments as lifelike we would be there since this is the objective.

Looking forward to your feedback on your pre-filtered files.

Philipp

 

[Robbo99999]

Thank you for this positive feedback. If you mean by live performance that you perceived the virtual instruments as lifelike we would be there since this is the objective.

Looking forward to your feedback on your pre-filtered files.

Philipp

I can't remember which one it was, but some of the pop type tracks literally felt like it was a live performance in a room, and recorded music doesn't normally sound like that unless it's recorded as live music. It felt like I could hear the room in a way. That's why I mentioned in my post a few times that it was best done with tracks the tester knows really well, because that way you can see if it's an effect that's been added to the music or not. But, yeah, it didn't feel like normal headphone listening, it felt like listening to music in a room, but I feel there might have been a bit of a side effect that it's making it sound like live music rather than recorded music. For instance my Anechoic Flat speakers don't make everything sound like live music in a room. Overall balance of the tonality didn't seem wrong though (if you were to assess the slope/trend of the music) - it didn't sound too bright & it didn't sound too dark (so at that level of assessment), but again would need to listen with tracks I know to see if I think each of the frequency response areas is being represented at roughly the right amount. I enjoyed checking it out, and will definitely check out the tracks you process for me, I'm looking forward to testing those and I'll try to do the listening & feedback to you as accurately & thoroughly as I can.

 

[fcserei]

Let me be a little crass here (with a strong IMHO added): stereo reproduction (not recording) sucks, and has nothing to do with lifelike reproduction. It produce a lot of conflicting auditory clues, confuses the brain and forces you to constantly evaluate the sound instead of enjoying it.
One of the many reasons is, that it does not and can not carry enough of the reverberant field, compared to the original venue where it was recorded. The reverberand sound is usually supressed, and a very little is usually mixed back in post production.

ASR now in the belief that if the spinorama of a speaker is right, then the room will magically rectreate the reverberant field what was omitted during recording in correct timing, spectrum and direction in normal stereo speaker reproduction. If you want to recreate what the producer/engineer heard, then it is an ok approach, because they faced the same problems and they had similar solutions.
If you want to recreate what the performance may have sounded in the original venue ( not in the control room), in my opinion there is only 2 ways with the existing library of recorded music: ambiophonics ( any kind of crosstalk cancellation with added ambience ) or binaural (hrtf binauralization with added ambience)

See what the common theme here: spectum, timing and direction correct recreated reverberant field.

Just hrtf manipulation can not reliably move the sound out of your head because it is just a small part of the complete set of auditory clues that you need for proper externalization. Put the performance in a believable space and add headtracking, and it is a gamechanger.

If the added ambience feels too echoy, it is because although the simulated anbience usually strong enough to mask the litle clues what are on the record, but they characteristics can clash, or the ambience power distribution is too frontal oriented instead of properly diffuse.

I usually try to match the ambience to the record: pick the correct type and size of concert hall for orchestral, Birdland or BlueNote for jazz, various opera houses for opera etc. Of coure it is a problem for music whic was never intended for live performance, but you can have very pleasant surprises with experiments.

 

[Robbo99999]

Let me be a little crass here (with a strong IMHO added): stereo reproduction (not recording) sucks, and has nothing to do with lifelike reproduction. It produce a lot of conflicting auditory clues, confuses the brain and forces you to constantly evaluate the sound instead of enjoying it.
One of the many reasons is, that it does not and can not carry enough of the reverberant field, compared to the original venue where it was recorded. The reverberand sound is usually supressed, and a very little is usually mixed back in post production.

ASR now in the belief that if the spinorama of a speaker is right, then the room will magically rectreate the reverberant field what was omitted during recording in correct timing, spectrum and direction in normal stereo speaker reproduction. If you want to recreate what the producer/engineer heard, then it is an ok approach, because they faced the same problems and they had similar solutions.
If you want to recreate what the performance may have sounded in the original venue ( not in the control room), in my opinion there is only 2 ways with the existing library of recorded music: ambiophonics ( any kind of crosstalk cancellation with added ambience ) or binaural (hrtf binauralization with added ambience)

See what the common theme here: spectum, timing and direction correct recreated reverberant field.

Just hrtf manipulation can not reliably move the sound out of your head because it is just a small part of the complete set of auditory clues that you need for proper externalization. Put the performance in a believable space and add headtracking, and it is a gamechanger.

If the added ambience feels too echoy, it is because although the simulated anbience usually strong enough to mask the litle clues what are on the record, but they characteristics can clash, or the ambience power distribution is too frontal oriented instead of properly diffuse.

I usually try to match the ambience to the record: pick the correct type and size of concert hall for orchestral, Birdland or BlueNote for jazz, various opera houses for opera etc. Of coure it is a problem for music whic was never intended for live performance, but you can have very pleasant surprises with experiments.

I'd just like to say that I think music reproduction should have nothing to do with making it sound like live music, it needs to sound as it has been recorded, and it's a creative construct from the artist & their engineers. Therefore, the goal of a playback system including this speaker virtualisation which Philipp has created shouldn't be about making it flavoured like it sounds like live music in church or wherever. I mean I suppose the conundrum is that he has to put some element of the room into the virtualisation in order to make it sound like speakers in a room, but it should really be a neutral room, and a good sounding room at that, and it would need to be a virtualisation of good speakers in such a room. I think it would be wrong to flavour it too much, and I think this would be the view of most ASR readers. It's important that the ambience of the recording is preserved rather than overwritten.

 

[myHRTF]

had noticed that you had warned users of the -10 dB preamp gain required, but I was wrongly under the impression that the bypass version would be 0 dB and personalized HRTF at -10dB, while it is actually the other way around. My bad. I wished I had the time to match to level match both versions, will try it again later.

Thank you, this is an important point: the -10 dB pre-amp gain requirement applies to the filters after purchase and download only. It does not apply to the demo songs on our website since this was factored in during creation of the demo songs.

In short, I was positively impressed how externalized was the rendition and it indeed felt like listening through speakers.

Thank you for this positive feedback.

My training was not very deep (undergraduate level), yet I am unsure why is it that HRTF almost always involve a simulation, while in my field, we would actually measure a few millimeters from the ear drum (with and without the ear canal probe) to get actual HRTF measurements. (Of note, I graduated, but never worked in the field, please excuse me in advance.)

This is a good point. In a nutshell: there is agreement among scientists that the HRTF should be measured close to the eardrum, unfortunately this is very uncomfortable and quickly becomes a serious health risk if not carried out by an expert with certified equipment. As an alternative, the blocked ear canal measurement was established, using capsule microphones with a fixture to keep it in place inside the entrance of the ear canal and to block it sufficiently. There are scientist who question the underlying theory of this method and others who question its measurement repeatability in practice. On the other hand, the majority of users got decent results with this method.

 

[myHRTF]

I tried the demo and I am positively surprised!

First of all, I have to thank you twice: I thank you for your time to go through the demo and write feedback, but even more I thank you for your contributions to the speaker virtualization community out there!

With myHRTF I was able to get a decent out of the head front localization, although the localization is not very accurate.

What was your feeling when choosing between the six ear shapes, could you clearly identify the best match or did you feel unsure about it? Reason I ask is that myself and others who tried our solution have a very accurate localization with pinpoint imaging. But it requires that the ear shape is set correctly. The described ear width measurement helps to narrow down the selection to 2-3 options, but the final decision should be made by listening carefully.

was missing a lot of the lower midrange

I have an idea what could be the reason for that and would love to try that with you. Would you be willing to repeat the demo with a special version?

May not produce the same accuracy as Smyth Realizer or Impulcifer, but is also a lot less hassle.

What puzzles me with these solutions is that they require a room with good acoustics for calibration first. Since this is difficult to get, most users end up with typical living room acoustics and more often than not this acoustics is flawed. In contrast, simulated room acoustics can be made flawless and allows for a degree of perfection that is unimaginable with real room acoustics.

 

[myHRTF]

If you want to recreate what the performance may have sounded in the original venue ( not in the control room), in my opinion there is only 2 ways with the existing library of recorded music: ambiophonics ( any kind of crosstalk cancellation with added ambience ) or binaural (hrtf binauralization with added ambience)

Agree with you, and this is the reason why we implemented both standard stereo and crosstalk-free stereo. On our website you can switch between them seamlessly in order to adapt the reproduction to the different music pieces. Obviously, with headphones you don't need to cancel crosstalk, you get crosstalk suppression in the order of 30 dB and more free of charge even with open models.

The classical music pieces even feature added ambience as pointed out.

 

[myHRTF]

Just finished uploading a new beta release that addresses most of the feedback received so far:

• Reduced early reflections for more accurate imaging / localization as pointed out by several. This also increases perceived proximity as pointed out by @olieb
• Slight increase in power level in the lower midrange as pointed out by @jaakkopasanen
• Slight decrease in power level in the treble range for some headphones as pointed out by @olieb

New features:

• Three speaker distances and room acoustics available now:
  • One meter (3.3 ft), dry room acoustics -> best suited for mixing and virtual reality / gaming.
  • Two meters (6.6 ft), live room acoustics -> best suited for music listening and TV / movies.
  • Three meters (10 ft), live room acoustics -> best suited for music listening.

You can test it online for free, you just need your headphones and a ruler / tape measure. Just go to:
myHRTF Personalized Filter Designer

PM me in case you would like to test this with your favorite songs.

The work on this new release delayed the addition of new headphone models as promised, but we are confident to get this done by end of this week.

Happy listening,

Philipp

 

[somebodyelse]

You are absolutely right. Next week we will upload more headphone models, including HD6XX, HD600 and HD650.

Philipp

Did these get forgotten in the update?

 

[myHRTF]

Did these get forgotten in the update?

Not forgotten, but delayed due to the work on the new release: we will add them in the coming days. Fortunately, this doesn't require uploading a new release since new headphone models can be added on the fly. I will update here as soon as this is done.

 

[Ostrol]

Hi Philipp,
First of all congratulations on the software: I have had a jaw-dropping moment listening to the music: it really does feel in front of me- for the first time since I have started on my "3d audio with headphones" journey. I have tried VSX (my favourite so far, especially with the new testing system), Lewitt space replicator and Virtuoso.
As for critiques to your beta:
I have a very big head and ear canals, and yet small ears - the ear measurement was extremely unintuitive. Cycling through the options the larger ear sizes gave me clean highs with no resonance peaks, yet with extremely woofy bass and bloated low-mids (the orchestra sample was extremely telling). Going for a small measurement fixed the lows but presented high frequency issues. A better explanation of the measurement process is needed, right now it seems to suggest 2 measurements (?) per ear. Should the ear canal be visible at point 1 and invisible at point 2? Quite confusing, at least for me.
I can't wait for this to be a fleshed out product! I would love to know how much you'd think to price this solution, also.

 

[somebodyelse]

A better explanation of the measurement process is needed, right now it seems to suggest 2 measurements (?) per ear. Should the ear canal be visible at point 1 and invisible at point 2? Quite confusing, at least for me.

I took it to mean one measurement with a different start point depending on the individual's ear shape (ear canal visible or hidden.) That we have different interpretations suggest some improvement is needed.

I tried it with the Moondrop Chu and found it better than the crossfeed plugins I've tried before in terms of getting the image out of head, but with a heavy frequency tilt boosting bass and cutting high end. I haven't experimented with changing to values that don't match my head/ear measurements though. I haven't tried any other HRTF models either - this has less of a barrier to entry than most.

 

[Robbo99999]

I took it to mean one measurement with a different start point depending on the individual's ear shape (ear canal visible or hidden.) That we have different interpretations suggest some improvement is needed.

I tried it with the Moondrop Chu and found it better than the crossfeed plugins I've tried before in terms of getting the image out of head, but with a heavy frequency tilt boosting bass and cutting high end. I haven't experimented with changing to values that don't match my head/ear measurements though. I haven't tried any other HRTF models either - this has less of a barrier to entry than most.

Yes, I did it the same way you did, that was my understanding.

 

[myHRTF]

First of all congratulations on the software: I have had a jaw-dropping moment listening to the music: it really does feel in front of me- for the first time since I have started on my "3d audio with headphones" journey.

Thank you for your time to test it and write feedback, and thank you for pointing this out.

Cycling through the options the larger ear sizes gave me clean highs with no resonance peaks, yet with extremely woofy bass and bloated low-mids

I'm puzzled a bit since there shouldn't be any difference in the bass and low midrange for the different ear shapes, it should only affect the high frequency range. With the ear shape you only set the perceived height of the front channels with the objective to have it at eye level. Would it make sense to try again with this information in mind?

Should the ear canal be visible at point 1 and invisible at point 2? Quite confusing, at least for me.

It is about the visibility of the ear canal as looked at from point 1 (perpendicular to your main head axis). The picture with point 1 and 2 indicate how to read the result from the ruler. Maybe this should be made more clear.
In any case, this measurement can only help to narrow down the number of suitable ear shapes from 6 to 2-3, the final decision should be made by listening. But for many people it helps to choose from only 2-3 options instead of 6.

I can't wait for this to be a fleshed out product! I would love to know how much you'd think to price this solution, also.

Prices and launch dates are still under discussion at this point in time, we will need a few more weeks to settle this.

 

[myHRTF]

I tried it with the Moondrop Chu and found it better than the crossfeed plugins I've tried before in terms of getting the image out of head

Thank you for your time to go through the demo and write feedback.

I took it to mean one measurement with a different start point depending on the individual's ear shape (ear canal visible or hidden.)

Your understanding is correct.

but with a heavy frequency tilt boosting bass and cutting high end.

To some extent this is normal since reflections contain less energy in the high frequency range. From your feedback I read that you would prefer less bass boost. Please correct me if I'm wrong.

I haven't experimented with changing to values that don't match my head/ear measurements though.

This is ok since these personalization settings are not supposed to be used as kind of EQ. Once set correctly, other settings decrease sound quality typically.