[71 dB]

Looks like I have been wrong about ITD. What I have said about it has been based on the Woodworth's formula: ITD = a/c * ( 𝞱+sin𝞱 ), but it never occured to me that it only applies to high frequencies (above 1 kHz or so). At lower frequences ITD is larger: Instead of 250 µs, about 400 µs corresponds 30° angle of sound. I apologize everyone for writing falsehoods about ITD. I really thought I had it right, but apparently that wasn't the case! Oh dear! Do I know/understand anything?

 

[gregorio]

Instead of 250 µs, about 400 µs corresponds 30° angle of sound.

Yes, providing the speakers are at 30° and the music only contains freqs at about 300Hz. If it contains 700Hz freqs, that corresponds to about 150µs ITD or 150Hz freqs correspond to about 500µs, it’s variable by freq! Plus, there’s still other factors associated with this single aspect; Those graphs represent an average of 130 people’s HRTFs, so it doesn’t show that each ear (of the same person) has a different HRTF/HRIR, with a different ITD vs freq curve and before you argue this doesn’t affect your enjoyment of music: Reliable evidence suggests this difference between the ITD of each ear plays a significant role in sound elevation discrimination. Crossfeed ignores all of this and a significant number of other factors and again, we’re just discussing the direct sound, reflections add another whole bunch of factors!

G

 

[71 dB]

Yes, providing the speakers are at 30° and the music only contains freqs at about 300Hz. If it contains 700Hz freqs, that corresponds to about 150µs ITD or 150Hz freqs correspond to about 500µs, it’s variable by freq! Plus, there’s still other factors associated with this single aspect; Those graphs represent an average of 130 people’s HRTFs, so it doesn’t show that each ear (of the same person) has a different HRTF/HRIR, with a different ITD vs freq curve and before you argue this doesn’t affect your enjoyment of music: Reliable evidence suggests this difference between the ITD of each ear plays a significant role in sound elevation discrimination. Crossfeed ignores all of this and a significant number of other factors and again, we’re just discussing the direct sound, reflections add another whole bunch of factors!

G

Yeah, now I know, but despite of all this, I genuinely do prever using crossfeed most of the time. It manages to do something important for me.

 

[gregorio]

Yeah, now I know, but despite of all this, I genuinely do prever using crossfeed most of the time.

I’ve never once questioned your perception or preference, just your explanation/theory of why crossfeed works. What crossfeed does is not natural or related to real life, with the single exception of ILD, it pretty much messes up everything. However, HPs without crossfeed is also pretty messed up because the mixes are primarily designed for speaker playback. So what we’re left with is two differently messed up HP presentations, how our personal perception responds to them and personal preference. There’s no scientific justification for preferring crossfeed beyond it just being how your personal perception responds to it. There is a scientific justification for a personalised set of HRTFs + reverb + head tracking though.

G

 

[71 dB]

I’ve never once questioned your perception or preference, just your explanation/theory of why crossfeed works. What crossfeed does is not natural or related to real life, with the single exception of ILD, it pretty much messes up everything. However, HPs without crossfeed is also pretty messed up because the mixes are primarily designed for speaker playback. So what we’re left with is two differently messed up HP presentations, how our personal perception responds to them and personal preference. There’s no scientific justification for preferring crossfeed beyond it just being how your personal perception responds to it. There is a scientific justification for a personalised set of HRTFs + reverb + head tracking though.

G

Unfortunately HRTFs + reverb + head tracking is not available for me, but crossfeed is.

 

[bigshot]

Yeah, now I know, but despite of all this, I genuinely do prever using crossfeed most of the time. It manages to do something important for me.

Perfectly fine. No argument there. Signal processing is a personal preference.

 

[sander99]

My HRTF was measured once

Unfortunately HRTFs + reverb + head tracking is not available for me, but crossfeed is.

What happened to your HRTF measurement results? They got lost? Or you didn't get access to them?

 

[71 dB]

Perfectly fine. No argument there. Signal processing is a personal preference.

Cool, but now there is the question of what am I doing on this discussion board? I can tell my preferences, but so what?

I have a lot of work ahead of me getting more familiar with the more precise aspects of ITD and how to implement them in my Nyquist plugins.

 

[71 dB]

What happened to your HRTF measurement results? They got lost? Or you didn't get access to them?

No access. My HRTFs are intellectual property of the Nokia Corporation.

 

[gregorio]

I have a lot of work ahead of me getting more familiar with the more precise aspects of ITD and how to implement them in my Nyquist plugins.

I think you’re flogging a dying horse, for a couple of reasons:

The “precise aspects of ITD” are far more complex than a simple static delay time time, as discussed it varies by freq but also, it has to be integrated with other factors, such as ILD obviously but also with the quite complex and variable freq response created by skull and pinnae diffraction and absorption. If you don’t do this, FR and ITD effectively fight with each other and can cancel each other out. Experiments show that a sound which should appear to be panned say hard left according to it’s ITD, will be perceived by some/many to be centrally panned if the skull/pinnae FR for a hard right panned sound is also applied. So if you ignore this or implement it with some simple static value, the results will be unpredictable from listener to listener and you’ll have done a lot of work but be back where you started. And, if you implement it correctly, then it’s no longer a crossfeed plug-in, it’s a HRTF plug-in!

However, HRTFs have a serious problem. While the processing power in mobile devices is now sufficient to cope with implementing them, the problem remains that generic HRTFs don’t work well for a lot of people and getting usable personalised HRTFs is very far from practical for consumers. However, there seems to be considerable research going on in this area. For example, there’s a published paper which uses simple skull measurements with a tape measure, correlates them with a database of HRTFs to come up with a function to modify a generic HRTF and thereby personalise it to a significant degree. However, I suspect most of the ongoing research isn’t being published because it’s being done in/by the corporate world. Apple filed a patent for using ear scans created by the iPhone’s “true depth” camera to personalise generic HRTFs. Dolby recently allowed personalised HRTFs to be used in their Atmos RMU (for content creators), implying something in the pipeline for consumers down the road. I would think that if Apple (and probably Dolby) are working on this, Google and some other big guns probably are too. So, I think we’ll see a lot more products incorporating HRTF/“spatial audio” functionality and incremental improvements to this generic HRTF problem, probably starting in the fairly near future.

G

 

[71 dB]

I think you’re flogging a dying horse, for a couple of reasons:

The “precise aspects of ITD” are far more complex than a simple static delay time time, as discussed it varies by freq but also, it has to be integrated with other factors, such as ILD obviously but also with the quite complex and variable freq response created by skull and pinnae diffraction and absorption. If you don’t do this, FR and ITD effectively fight with each other and can cancel each other out. Experiments show that a sound which should appear to be panned say hard left according to it’s ITD, will be perceived by some/many to be centrally panned if the skull/pinnae FR for a hard right panned sound is also applied. So if you ignore this or implement it with some simple static value, the results will be unpredictable from listener to listener and you’ll have done a lot of work but be back where you started. And, if you implement it correctly, then it’s no longer a crossfeed plug-in, it’s a HRTF plug-in!

However, HRTFs have a serious problem. While the processing power in mobile devices is now sufficient to cope with implementing them, the problem remains that generic HRTFs don’t work well for a lot of people and getting usable personalised HRTFs is very far from practical for consumers. However, there seems to be considerable research going on in this area. For example, there’s a published paper which uses simple skull measurements with a tape measure, correlates them with a database of HRTFs to come up with a function to modify a generic HRTF and thereby personalise it to a significant degree. However, I suspect most of the ongoing research isn’t being published because it’s being done in/by the corporate world. Apple filed a patent for using ear scans created by the iPhone’s “true depth” camera to personalise generic HRTFs. Dolby recently allowed personalised HRTFs to be used in their Atmos RMU (for content creators), implying something in the pipeline for consumers down the road. I would think that if Apple (and probably Dolby) are working on this, Google and some other big guns probably are too. So, I think we’ll see a lot more products incorporating HRTF/“spatial audio” functionality and incremental improvements to this generic HRTF problem, probably starting in the fairly near future.

G

Well, my plugins are for music making. I have used my simple ITD and ILD models for that, but now I can, with some effort, make those models more accurate. I don't deal with dead horses. I have my own young pony. It is far from dead.

I don't amplitude pan, I use the combination if ILD and ITD to come up with panning that works better for headphones.

 

[Atriya]

I think you’re flogging a dying horse, for a couple of reasons:

The “precise aspects of ITD” are far more complex than a simple static delay time time, as discussed it varies by freq but also, it has to be integrated with other factors, such as ILD obviously but also with the quite complex and variable freq response created by skull and pinnae diffraction and absorption. If you don’t do this, FR and ITD effectively fight with each other and can cancel each other out. Experiments show that a sound which should appear to be panned say hard left according to it’s ITD, will be perceived by some/many to be centrally panned if the skull/pinnae FR for a hard right panned sound is also applied. So if you ignore this or implement it with some simple static value, the results will be unpredictable from listener to listener and you’ll have done a lot of work but be back where you started. And, if you implement it correctly, then it’s no longer a crossfeed plug-in, it’s a HRTF plug-in!

However, HRTFs have a serious problem. While the processing power in mobile devices is now sufficient to cope with implementing them, the problem remains that generic HRTFs don’t work well for a lot of people and getting usable personalised HRTFs is very far from practical for consumers. However, there seems to be considerable research going on in this area. For example, there’s a published paper which uses simple skull measurements with a tape measure, correlates them with a database of HRTFs to come up with a function to modify a generic HRTF and thereby personalise it to a significant degree. However, I suspect most of the ongoing research isn’t being published because it’s being done in/by the corporate world. Apple filed a patent for using ear scans created by the iPhone’s “true depth” camera to personalise generic HRTFs. Dolby recently allowed personalised HRTFs to be used in their Atmos RMU (for content creators), implying something in the pipeline for consumers down the road. I would think that if Apple (and probably Dolby) are working on this, Google and some other big guns probably are too. So, I think we’ll see a lot more products incorporating HRTF/“spatial audio” functionality and incremental improvements to this generic HRTF problem, probably starting in the fairly near future.

G

I wonder if the paper you're referring to is this one:

https://drive.google.com/file/d/1JVDNxQreYzg7jfauMwFnK3Sg21aB5ri5/view?pli=1

(This one is actually a dissertation.)

Most of it isn't written in English, but they do have a free software that I used to generate a .sofa file from head measurements:

https://github.com/davircarvalho/Individualized_HRTF_Synthesis/tree/master/Individualized HRTF App

Plugging in that .sofa file into APL Virtuoso, I seem to get better results than the generic HRTFs than come with Virtuoso.

 

[gregorio]

I wonder if the paper you're referring to is this one:

https://drive.google.com/file/d/1JVDNxQreYzg7jfauMwFnK3Sg21aB5ri5/view?pli=1

I don’t recall exactly the paper I was referring to, I would probably have linked to it if I had, but I’m sure it’s not the one you suggest because I don’t read Spanish.

G

 

[71 dB]

I don’t read Spanish.

G

Apparently you don't read Portuguese either, or you aren't aware of the fact that they speak Portuguese in Brazil.

 

[Atriya]

Apparently you don't read Portuguese either, or you aren't aware of the fact that they speak Portuguese in Brazil.

Haha, this is exactly why I wrote "Most of it isn't written in English" instead of naming a language, since I thought it might be Portuguese, but didn't want to embarrass myself just in case it was Spanish (or something else).