[cachi]

If i was rich i would have invested in the companies like Smyth Research. I truly believe the A16 can have multiple use and will really revolutionise the sound movies market for home use.
If they would invest more into advertising with magazines like What HiFi, then everything will be at a different level by now, however let's have more patience, is still January.

I think the future of personalized audio will entail a much better solution than the 'smyth' solution (albeit its a very good solution currently). I bet that it's possible to create a one-to-one map between ear photo/scan and HRTF. I have a good idea about how to do it but don't have the time or the interest to dedicate myself to it. I'm sure someone will do it eventually though since its kind of an obvious solution to go for

 

[jgazal]

I have a good idea about how to do it but don't have the time or the interest to dedicate myself to it.

Please describe your idea. I am genuinely curious to know alternative solutions.

 

[cachi]

Please describe your idea. I am genuinely curious to know alternative solutions.

Well, i'm not sure if your familiar with the concept of a basis set (like in linear algebra) but basically it would entail deconstructing ear shapes into a good basis set. Say maybe like a flat plane, a curved surface (3d sin wave perhaps), and other basic surface shapes. Then measure the HRTF for this 'basis' set. A good basis set would allow you to recreate all ears by summing the weights of the basic units which best math the user's ear morphology. For instance, say your ears concha requires different weights for curved basis functions than mine, hence personalized HRTF would be possible by scanning or taking a picture of an individuals ear and fiiting the basis functions

 

[jgazal]

Well, i'm not sure if your familiar with the concept of a basis set (like in linear algebra) but basically it would entail deconstructing ear shapes into a good basis set. Say maybe like a flat plane, a curved surface (3d sin wave perhaps), and other basic surface shapes. Then measure the HRTF for this 'basis' set. A good basis set would allow you to recreate all ears by summing the weights of the basic units which best math the user's ear morphology. For instance, say your ears concha requires different weights for curved basis functions than mine, hence personalized HRTF would be possible by scanning or taking a picture of an individuals ear and fiiting the basis functions

You should join the industry with your efforts:

Everyone in this industry is working on it. Myself included. The team at Dirac know how do rooms very well. Smyth brothers have been doing this far longer and have
years of "trade secrets" they founded DTS. They also wrote the code for APTx.

Qualcomm is working of image based HRTF. Along with other production tools.

Fraunhofer is on the same path.

Dolby is working on atmos for headphones

All the university's in audio research are working on HRTF related theories.

Everyone has the same issues. The human!

I believe the more brilliant people working on this the better.

 

[pfzar]

Stay tuned.......... something coming.

 

[cachi]

You should join the industry with your efforts:



I believe the more brilliant people working on this the better.

Actually, you know what. The realiser might be a good tool to develop this idea if it would let me extract the measured HRTF from it. Since I could then measure different basic shapes to try and come up with a basis set. Once I have that then I could use my own ears to try to recreate my HRTF with the basis set. I know the iphone X has a pretty cool point cloud scanner, so it should be possible with the future of phones to get accurate scans of people ears

 

[pfzar]

The Smyth réaliser is capturing the BRIR.. In order to personalize a HRTF you should not be capturing a room. You add a room to the tail of an anechoic HRTF. Then convolve with coordinates of the intended speaker location.. You also can't extract the BRIR's from the Smyth, its stored as binary data sets.

Most HRTF's are captured at 1024 data points per ear (2048). The Smyth is capturing 16per ear. (max 32). Not enough data to extrapolate or interpolate all the possible points in a sphere.

BRIR=Binaural Room Impulse Response=A head in a room with loudspeakers, measuring interaction of the room and head.
HRIR=Hear Related Impulse Response=A head in an anechoic room with loudspeakers, measuring the interaction of speaker location and head.

 

[jgazal]

(...)
Most HRTF's are captured at 1024 data points per ear (2048). The Smyth is capturing 16per ear. (max 32). Not enough data to extrapolate or interpolate all the possible points in a sphere.
(...)

What do you think about the Realiser interpolation performance with an Atmos virtual speaker arrangement? Did it sound convincing to you?

 

[pfzar]

I heard it in Denmark about 2 years ago, Its quite convincing. But you can see the speakers so your brain is a little biased.

 

[jgazal]

I heard it in Denmark about 2 years ago, Its quite convincing. But you can see the speakers so your brain is a little biased.

Interesting.

So what is your opinion about its interpolation performance without speakers at sight?

I tried to ask them about stereoscopic pictures of the measured room, but perhaps the VR headsets are just not developed enough for such task...

 

[jgazal]

@pfzar, I am very interested in using a sound virtualization DSP with a VR headset and Netflix VR:

More examples of attempts on immersive audio:

(...)

3. Netflix VR

Probably atmos bed and objects downmixed to binaural with generic HRTF. Potentially better with a Realiser crossfeed free PRIR.

​

(...)

Do you recommend Netflix to add virtual speakers to their virtual room visual rendering?

I am eager to discover the solution THX will adopt to mitigate such “speaker at sight restriction”, if it is really a relevant restriction.

Are you planning some sort of head tracking with a high density HRIR? Are you planning to use a generic or a personalized HRIR?

Do you believe that HRIR density is a worse restriction than having personalized instead of a generic HRIR or BRIR?

I am very curious to test!

 

[sander99]

I have some questions, and a few remarks.

Most HRTF's are captured at 1024 data points per ear (2048)

Does that mean you measure the HRTF for 1024 directions, and that for each ear? For the moment I assume that is what it means.
By the way: what is the maximum angle between 2 adjacent points in that case?

The Smyth is capturing 16per ear. (max 32).

Actually if it measures 16 speakers at 5 measurement points (5 lookangles to interpolate between for headtracking/derotating), that would be 80 per ear, 160 total.
Also for one individual pair (one speaker, one lookangle, for the 2 ears) beware that in fact that for that specific lookangle the full HRTF (all directions, an infinite number) is respected: reflexions and reverberations from any direction are captured including the HRTF filtering they underwent. This holds for all the speakers, and all the measurement points. So when during the simulation you look for example straight forward (the zero degree measurement point), at that moment the simulation of all the speakers and the room is "accurate" in the sense that it is not in anyway compromised by lack of having a 1024/2048 point HRTF. Only at the moment that you move your head inbetween the measurement points and the Realiser starts interpolating then there could be (and probably will be) inaccuracy in that same sense.

The Smyth réaliser is capturing the BRIR.

A little detail: When you do a personal measurement with the Realiser you are actually capturing a PRIR (Personalised Room Impulse Response), not a BRIR (although it would be a BRIR for someone else). Indeed you can not extract the personal HRTF from that, but the personal HRTF is fully respected in the sense that I descibed above.

Note that in the approach of capturing an isolated HRIR and combining that with an isolated room response later, to make that "accurate" you should somehow make sure that all refexions and reverberations undergo the proper HRTF filtering belonging to the many different directions they are approaching the ears from. Is this incorporated in what you desribe in the following quote?

You add a room to the tail of an anechoic HRTF. Then convolve with coordinates of the intended speaker location.

 

[pfzar]

Sorry formatting got weird. I will post a reply when I get back to my computer

 

[MichaelJames99]

So, while we wait, I have nothing that does surround processing for movies from my Bluray or DirectTV.. anything fairly cheap that I can buy to make due until this gets done? Something under $500
Maybe the creative sounds X7?

 

[cachi]

So, while we wait, I have nothing that does surround processing for movies from my Bluray or DirectTV.. anything fairly cheap that I can buy to make due until this gets done? Something under $500
Maybe the creative sounds X7?

Have you checked out the sennheiser gx1000 headphone processor? There are some binaural demos of it on youtube. Its $200