immersifi
100+ Head-Fier
Uli87 said:
"Wow thanks for all this info. I found you also posting on gearslutz talking to a guy about Holophnics and Zuccarelli when doing a google search.
I found a really cool video on BRIR on youtube https://www.youtube.com/watch?v=bmOvHjSBlnc.
These sort of technologies are already implemented into recording and mixing programs like protools? Because I know of some pretty awesome music that was recorded on smaller budgets in small home studios that also manage to capture pretty great depth. I mean like when you pan left and right while mixing, is there a way to actually add depth on a z axis?"
[sorry for the 'butchered' quote - I meant to quote you, but typed my reply first...and without coffee...Now, my reply]
First, you're most welcome.
Second, Yes.
ADDED : In the BRIR video cited by Uli87, the mannequin head shown right around the 1:50-mark looks like it is the Neumann KU-100, which is a well-known diffuse field-equalized mannequin mic - it's also the same type that is used in all of my (immersifi recording services) work.
If you canvass the 'web, you will find that people like me (acousticians, recording engineers, etc.) often post impulse responses - they may be monaural, they may be stereo, they may be binaural etc. I can't recall the links this moment, but basically, there are sites wherein they will tell you the format, type of mic (so you can de-convolve the mic's frequency response if so desired).
The thing is this... the realism of the reverb generated relies upon a few things, but most notably the quality of the impulse response, and there are several things that govern the accuracy of that part of the puzzle - all of that is way beyond the scope of this post. However, a good place to start is to review the lectures of Dottore Angelo Farina (University of Parma).. There are so many issues surrounding the 'accuracy' of an impulse response. My point here is that while there are free impulse responses all over the web, like most things...'caveat emptor' applies.
As far as obtaining the IR (whether BRIR or others), there are a few typical means of gathering the IR, and each have their pro's and con's; some are better suited to noisy spaces (MLS-based (circular cross-correlation)) others are better suited at de-convolving the non-linearity of the speaker used to generate the impulse (Farina's method of the chirp and de-convolution based on time-reversed kernel), and so on. See, the challenge with all IRs is that you are often up against dynamic range constraints, extraneous noises, speaker non-linearities and such, and so there are things that need to be considered. Delta's are the desired source, but actually making a true delta function is a bit tricky, however, as Farina has pointed out, if your main area of interest is the HF portion of the IR, then remarkably, firecrackers are pretty repeatable (not line-on-line, but not bad) as ersatz delta functions. Popped balloons is another way to get there, but the size, pressure inside of, and shape of the balloon affect its spectral composition when it bursts.
As far as the elevation question you pose...clearly in VR, if they have a database of IRs as a function of X, Y, and Z (and of rotation about those axes - particularly critical for BRIR) and can interpolate to the next valid IR if there are a wealth of points.
That is, it's impossible to have every single IR as a function of mannequin head mic location and orientation, however, if 'n' is sufficiently large, then a convolution algorithm can sort of interpolate between the individual IRs as the listener turns his or her head. Likewise, the more distinct the IRs are, the more of them you need to make the interpolation not noticeable in real time. On the otehr hand, if you had a truly diffuse field (a sanctioned reverb chamber), then the IRs would not be all that different from location to location within the reverb chamber. Likewise for free-field conditions - in true free field conditions (no reflections), everything would simply come down to magnitude and orientation with respect to the source. So...those two (free and diffuse) represent the ends of the continuum where things are well-defined, but in reality, all other spaces in which we live, breathe, and experience music occupy a point on that continuum, and this, the IRs as a function of source-to-receiver distance, orientation and the like are more and more disparate.
As far as hearing how convolution using BRIRs can be used to design the acoustics of spaces...you should check out this demo from ODEON - seriously, this diversion is wholly worth taking...some very cool demos here, and at the core is the concept of the BRIR.
http://www.odeon.dk/auralisation
This code is written around ray-tracing and models of a space (virtual or real) in which the material properties are described (for each and every surface), and then BRIRs are used so that dry sounds, recorded in an anechoic chamber, can be convolved with the BRIR, itself being the result of a simulation of the ray-tracing and the location in the virtual space.
Great stuff.
Another similar bit of code is CATT.
Mark (immersifi)
"Wow thanks for all this info. I found you also posting on gearslutz talking to a guy about Holophnics and Zuccarelli when doing a google search.
I found a really cool video on BRIR on youtube https://www.youtube.com/watch?v=bmOvHjSBlnc.
These sort of technologies are already implemented into recording and mixing programs like protools? Because I know of some pretty awesome music that was recorded on smaller budgets in small home studios that also manage to capture pretty great depth. I mean like when you pan left and right while mixing, is there a way to actually add depth on a z axis?"
[sorry for the 'butchered' quote - I meant to quote you, but typed my reply first...and without coffee...Now, my reply]
First, you're most welcome.
Second, Yes.
ADDED : In the BRIR video cited by Uli87, the mannequin head shown right around the 1:50-mark looks like it is the Neumann KU-100, which is a well-known diffuse field-equalized mannequin mic - it's also the same type that is used in all of my (immersifi recording services) work.
If you canvass the 'web, you will find that people like me (acousticians, recording engineers, etc.) often post impulse responses - they may be monaural, they may be stereo, they may be binaural etc. I can't recall the links this moment, but basically, there are sites wherein they will tell you the format, type of mic (so you can de-convolve the mic's frequency response if so desired).
The thing is this... the realism of the reverb generated relies upon a few things, but most notably the quality of the impulse response, and there are several things that govern the accuracy of that part of the puzzle - all of that is way beyond the scope of this post. However, a good place to start is to review the lectures of Dottore Angelo Farina (University of Parma).. There are so many issues surrounding the 'accuracy' of an impulse response. My point here is that while there are free impulse responses all over the web, like most things...'caveat emptor' applies.
As far as obtaining the IR (whether BRIR or others), there are a few typical means of gathering the IR, and each have their pro's and con's; some are better suited to noisy spaces (MLS-based (circular cross-correlation)) others are better suited at de-convolving the non-linearity of the speaker used to generate the impulse (Farina's method of the chirp and de-convolution based on time-reversed kernel), and so on. See, the challenge with all IRs is that you are often up against dynamic range constraints, extraneous noises, speaker non-linearities and such, and so there are things that need to be considered. Delta's are the desired source, but actually making a true delta function is a bit tricky, however, as Farina has pointed out, if your main area of interest is the HF portion of the IR, then remarkably, firecrackers are pretty repeatable (not line-on-line, but not bad) as ersatz delta functions. Popped balloons is another way to get there, but the size, pressure inside of, and shape of the balloon affect its spectral composition when it bursts.
As far as the elevation question you pose...clearly in VR, if they have a database of IRs as a function of X, Y, and Z (and of rotation about those axes - particularly critical for BRIR) and can interpolate to the next valid IR if there are a wealth of points.
That is, it's impossible to have every single IR as a function of mannequin head mic location and orientation, however, if 'n' is sufficiently large, then a convolution algorithm can sort of interpolate between the individual IRs as the listener turns his or her head. Likewise, the more distinct the IRs are, the more of them you need to make the interpolation not noticeable in real time. On the otehr hand, if you had a truly diffuse field (a sanctioned reverb chamber), then the IRs would not be all that different from location to location within the reverb chamber. Likewise for free-field conditions - in true free field conditions (no reflections), everything would simply come down to magnitude and orientation with respect to the source. So...those two (free and diffuse) represent the ends of the continuum where things are well-defined, but in reality, all other spaces in which we live, breathe, and experience music occupy a point on that continuum, and this, the IRs as a function of source-to-receiver distance, orientation and the like are more and more disparate.
As far as hearing how convolution using BRIRs can be used to design the acoustics of spaces...you should check out this demo from ODEON - seriously, this diversion is wholly worth taking...some very cool demos here, and at the core is the concept of the BRIR.
http://www.odeon.dk/auralisation
This code is written around ray-tracing and models of a space (virtual or real) in which the material properties are described (for each and every surface), and then BRIRs are used so that dry sounds, recorded in an anechoic chamber, can be convolved with the BRIR, itself being the result of a simulation of the ray-tracing and the location in the virtual space.
Great stuff.
Another similar bit of code is CATT.
Mark (immersifi)
