[linux-audio-user] Audio 3-D Demo --- Any Interest in Software?

davidrclark at earthlink.net davidrclark at earthlink.net
Tue Jan 6 12:37:49 EST 2004


Thanks for your great mailing.  You've raised some excellent points.

>Will IR tell me that the average/desirable Sabin value from 20Hz to 
4KHz for a .3 reverberation time in a 17X14X7 room is 148.6?  

>A fundamental problem I see with IR is that the accoustical values of 
good rooms are being applied to signals that are created in very poor 
accoustical environments where frequency responses in the audible 
range are not flat. In this scenario, only half the problem is being 
addressed. This might not be a problem with IR as much as it is a mis-
use or underutilization issue.

You are absolutely right on both points.  This is why I emphasized 
earlier that the application of the impulse response function is the 
"trivial" part of the problem.  The larger problem is to design either 
a real room or a virtual room for the sound.  It's also why I asked 
about instrument generation.  Physical modelling of both the 
instruments and the room is one way to provide the non-acoustical 
sources Steve mentioned in an idealized environment.  [*]

It's more of a misuse and underutilization issue, IMO, as I'll discuss 
later in this mailing.

>My point is that great room modeling, reverberation and echo tools 
are "the cart before the horse."

In some cases, that's correct; in others it's not correct.  Many of us 
have sampler synths or sample CD's or purely synthesized tones (and I 
have my instrument generator).  Many of these samples are fine for 
putting into a room modelling program.  

But --- I certainly do agree wholeheartedly that discussing IR's is 
premature if one isn't thinking along the lines you've mentioned in 
your mailing.  IR's are way down the line, practically the last step 
as they are actually applied.  But the considerations that lead to 
them are present throughout the recording and mixing process.

>Almost everyone on this list has a recording studio in an untuned 
room.  [Snip] So, what good does it do to put that poorly recorded 
source into a great room?

>I'd suggest that the first thing to deal with is correcting these bad 
rooms with Virtual Rooms. You know, crap in equals crap out.

I'm not sure whay you mean by "Virtual Room" in this context.  My 
program can be used to create virtual rooms.  The way I handle bad 
recordings (after trying not to do that) is to superimpose a frequency 
spectrum on top of the frequency response of the virtual room that I 
created in my 3-D program.  In other words, as long as it's a linear 
process (as Steve said), you can combine everything into the same 
frequency response, then obtain a single impulse response function 
that both corrects any tuning type of problem (as well as whatever 
else you want to throw in) and models a 3-D environment complete with 
a frequency-dependent decay.  It's all part of the same process of 
getting the total environment set up correctly.  You then look at test 
signals recorded in your non-optimum setting, passed through the 
wonderful IR you've created, and everything sounds --- uh, sort of OK.

Another thing one can do is to model the crappy room backwards in 
time.  You would ask yourself, "What crappy signal do I need to create 
at the source to produce a good signal at location (x,y,z)?"  The 
impulse response function for that case can be generated, and it can 
be used to "correct" problems.  Now not all problems can be corrected, 
but certainly incorrect frequency responses can be.  This is what one 
is actually doing with a multi-band equalizer.

Another consideration that many audio engineers overlook is that most 
people listen to most music in crappy listening environments --- like 
MP3's with cheap headphones or in a car with the A/C going.  Most of 
these crappy listening environments completely swamp out everything 
that the audio engineers did to make the sound as good as it can be.  
But there is a possible solution to that: To some extent, a known 
crappy environment such as the typical American living room can be 
somewhat corrected by the backwards-in-time (retrodiction) room 
modelling.  Admittedly, there are a number of problems that cannot
be corrected, but:

Room modelling has a number of applications, not just determining the
frequency response of an ideal concert hall.
>The question I can't answer is, can Virtual Rooms sounds as good as 
physical rooms that are tuned?

That's a multimillion dollar question, isn't it?  I'd like to find 
out, which is the real reason why I created these programs in the 
first place.  So far, it sounds even better than I had hoped for.


[*] The ability to model both instruments and the environment leads to 
the "rethinking" of the entire process that I referred to in an 
earlier mailing.  If you can model the instruments, then you don't 
necessarily need to play them in the time sequence of a recording.  
"Recording" of the audio doesn't exist any more.  It's more of an 
assembly or compilation process.  However, artists need not worry 
because one still needs to perform the score which does need to be 
recorded somehow for a realistic performance.

More information about the Linux-audio-user mailing list