[RRod]

Unfortunately it seems that Audeze Mobius doesn't offer the same level of HRTF customization as Creative Super X-Fi does. Ticks all the other boxes though.

Btw I'm now so convinced about Creatives new thing that I decided to drop my own project for the neural network based HRTF approximator. I don't think I could have achieved better than what they promise to deliver from the pictures.

Why not do it anyway and see how well you do? What database are you using to get your anthro. and impulse data, btw?

Of the current crop of virtualization products, the SXFi also gives me the most excitements. I have no doubt the Realiser will work like a screwin' champ after I've found a s̶u̶c̶k̶e̶r̶ nice person to let me measure myself in their Atmos setup. But the real trick is NOT forcing the user to sit with binaural mics in their ears.

 

[jaakkopasanen]

Why not do it anyway and see how well you do? What database are you using to get your anthro. and impulse data, btw?

Of the current crop of virtualization products, the SXFi also gives me the most excitements. I have no doubt the Realiser will work like a screwin' champ after I've found a s̶u̶c̶k̶e̶r̶ nice person to let me measure myself in their Atmos setup. But the real trick is NOT forcing the user to sit with binaural mics in their ears.

I was planning to gather the data myself by measuring bunch (hopefully 100-200) of people with binaural microphones in a movie mixing studio. End users would be presented with two of these PRIRs of which the user selects the one that sound more realistic and then another pair is presented and another. This way I would get information about similarities of different PRIRs which would make it possible to frame them in a multidimensional vector space. Each end user has her own HRTF which is also located somewhere in that vector space. The task would be to find the user's vector and select the one PRIR that is closest to it in the vector space. I also thought about generating the impulse responses from the vector representations with recurrent neural network which would make it possible to obtain PRIR that matches the vector instead of selecting the closest one.

So way not do it anyways? Because it's a lot of work and would probably cost me quite a lot of money. I was prepared to do it because I want to build awesome stuff and a solution which offers easy HRTF approximation doesn't exist, or at least until Creative's Super X-Fi. However I still do plan to do the measurements for myself and possibly to couple of other people as well. Use them with HeSuVi on Windows and Viper4Android on Android.

 

[RRod]

I was planning to gather the data myself by measuring bunch (hopefully 100-200) of people with binaural microphones in a movie mixing studio. End users would be presented with two of these PRIRs of which the user selects the one that sound more realistic and then another pair is presented and another. This way I would get information about similarities of different PRIRs which would make it possible to frame them in a multidimensional vector space. Each end user has her own HRTF which is also located somewhere in that vector space. The task would be to find the user's vector and select the one PRIR that is closest to it in the vector space. I also thought about generating the impulse responses from the vector representations with recurrent neural network which would make it possible to obtain PRIR that matches the vector instead of selecting the closest one.

So way not do it anyways? Because it's a lot of work and would probably cost me quite a lot of money. I was prepared to do it because I want to build awesome stuff and a solution which offers easy HRTF approximation doesn't exist, or at least until Creative's Super X-Fi. However I still do plan to do the measurements for myself and possibly to couple of other people as well. Use them with HeSuVi on Windows and Viper4Android on Android.

Ah I see, I thought perhaps you were starting with something like the IRCAM HRTF data. Yes, data collection is beaucoup time and money, so I don't blame you for waiting out for how well the SXFi does. Just to understand, though, what would your ultimate user-end features be for prediction? Is the goal to make a decent prediction based upon a few selections of preference among the PRIRs?

 

[jaakkopasanen]

Ah I see, I thought perhaps you were starting with something like the IRCAM HRTF data. Yes, data collection is beaucoup time and money, so I don't blame you for waiting out for how well the SXFi does. Just to understand, though, what would your ultimate user-end features be for prediction? Is the goal to make a decent prediction based upon a few selections of preference among the PRIRs?

Yes. User does a few selection after which the PRIR is generated. HRTF approximation becomes more accurate after each selection until a final indivisible region in the vector space has been found. They are binary selections so accuracy would largely depend on the number of measured PRIRs.

 

[RRod]

Yes. User does a few selection after which the PRIR is generated. HRTF approximation becomes more accurate after each selection until a final indivisible region in the vector space has been found. They are binary selections so accuracy would largely depend on the number of measured PRIRs.

I would have found such a solution useful when I was trying to pick and optimal filter for OpenAL (which lets you import HRTF info in a certain form). The hard part was going through all ~50 options from IRCAM and doing an honest comparison, so your method would have been a perfect fit.

 

[castleofargh]

I would have found such a solution useful when I was trying to pick and optimal filter for OpenAL (which lets you import HRTF info in a certain form). The hard part was going through all ~50 options from IRCAM and doing an honest comparison, so your method would have been a perfect fit.

for having done just that, I +1 that comment. real PITA to test them one by one. although as I mentioned somewhere already, I rapidly noticed some correlation between the size of the head and how well it seemed to work for me (at least in some key aspects of the imaging). accidental or real, if I went at it again someday, I'd save time by rejecting the smaller heads(when the information is available).

 

[jaakkopasanen]

I would have found such a solution useful when I was trying to pick and optimal filter for OpenAL (which lets you import HRTF info in a certain form). The hard part was going through all ~50 options from IRCAM and doing an honest comparison, so your method would have been a perfect fit.

Now that I think about it the idea of collecting my own HRIR database seems very silly. Maybe I'll look into the IRCAM database at some point. At least build an UI for these binary selections and generate HRIR files fit for HeSuVi. No promises though, I'm busy enough with my real business.

Another way how headphone audio can improve: equalization profiles for all the headphone models. I'm actually working on this too. There's a thread about it here in sound science: https://www.head-fi.org/threads/how-to-make-equalization-sound-good.873773/ and the project repo is in GitHub: https://github.com/jaakkopasanen/AutoEq. Of course there's Sonarworks' TrueFi but that only has some 120 headphone models, missing for example my own HE-400S.

 

[Glmoneydawg]

Now that I think about it the idea of collecting my own HRIR database seems very silly. Maybe I'll look into the IRCAM database at some point. At least build an UI for these binary selections and generate HRIR files fit for HeSuVi. No promises though, I'm busy enough with my real business.

Another way how headphone audio can improve: equalization profiles for all the headphone models. I'm actually working on this too. There's a thread about it here in sound science: https://www.head-fi.org/threads/how-to-make-equalization-sound-good.873773/ and the project repo is in GitHub: https://github.com/jaakkopasanen/AutoEq. Of course there's Sonarworks' TrueFi but that only has some 120 headphone models, missing for example my own HE-400S.

My biggest problem that needs adressing at this very moment is.....i can hear my habernero honey peanuts crunching in my head!...noise cancellation perhaps?

 

[bigshot]

That won't work. I just got a set of noise cancelling headphones. They don't make the voices in my head any quieter!

 

[Glmoneydawg]

That won't work. I just got a set of noise cancelling headphones. They don't make the voices in my head any quieter!

Well thats disappointing!

 

[71 dB]

I just got a set of noise cancelling headphones. They don't make the voices in my head any quieter!

Tinnitus-canceling headphones! That would be something!

 

[Glmoneydawg]

That won't work. I just got a set of noise cancelling headphones. They don't make the voices in my head any quieter!

Hmmm...i see an untapped market here.

 

[stalepie]

 

[skwoodwiva]

Tinnitus-canceling headphones! That would be something!

My wife has it. They call it "the marrage breaker".
Sad I am, 8 mo now. Indeed it is its definition.
Just searching you out...

 

[skwoodwiva]

Bigger magnets, anyone?
They are here!