Rational Acoustics



phasetransitions
August 7th, 2008, 08:44 PM
Hello world,

I posted this as a comment on Dave Stagl's blog (www.goingto11.com), and I thought that it would make a nice first post to the Rational forum, since it ended up being pretty long. Its my thought's on Dave's blog post of non "flat" system tunings:

Dave,

My thoughts on “non-flat” eq curves below. But first, its important to understand where the X-curve came from. The X-curve (curves really) was a result of trying to compensate for the additional high frequency absorption in the reverberant field of large rooms, when measuring the in room response with a time-blind RTA analyzer.

Since the integration time constant of the RTA was always long enough to include reverberant field effects at all frequencies, the natural tendency of large rooms to have more air and/or wall absorption, or diffusion events leading to absorption, the in room response with a flat RTA curve was far too bright.

Its also important to remember that the x-curve is also applied in the mixdown space, as well as the playback space, so in a sense is completely factored out of the mix process. I think the X-curve was an elegant solution for the measurement limitation of the day, but it is no longer really necessary.

—-

Now, what are my thoughts on non-flat equalization curves? Again, this is not a simple process, because with the x-curve above, the degree to which the room influences the balance is going to depend on the nature of the measurement system involved. With modern analyzers, you can look at the time domain, and include as little or as much as you feel appropriate.

In the case of Smaartlive, the fixed point per octave (FPPO) process is elegant because it provides short windows at high frequencies, essentially hiding the room’s absorption effects, but uses longer windows in the mids and lows, including more of the room effects. So a “flat” curve in the typical Smaartlive situation is excluding the HF absorption effects that plagued the RTA analyzer used with the x-curve.

Now, if you are still along for the ride, the power response of real loudspeakers is not constant with frequency, because their directivity is not constant with frequency. For the on axis response of a system to have balance, the low and mid power response must be greater, because a lot of that energy is spilling around the box, and radiating out in directions other than the axial direction. So, if you were to apply heavy time windowing to the low end response of the loudspeaker, essentially only looking at the direct field response, and THEN only look at the axial response, you could end up with a substantially low/mid tilted global frequency response balance.

The FPPO response windowing in Smaartlive includes more of the room in the range where the room’s acoustic behavior matters the most, and in the range where real-world audio systems have less forward directivity control. This, I think was, Sam B’s best insight when he created Smaart.

—-

So, with discussion of how the measurement system will influence the perception of the total system transfer function, back to the original question of the non-linear system response curve. One place that I feel the system response should almost always be “non-linear” is in the range below about 100-120hz. Sound systems tailored for an extra 6-10dB of output in the lows are almost always appropriate for modern mixing. The trick to this extra low end is to have it smoothly transition over the 60-120hz octave, so it does not become boomy or muddy.

Now, for the rest of spectrum, I think the system response behavior can be tailored towards “very linear” at low to moderate volume levels. If the system is going to be used at 95dBA or below, I prefer to leave a linear transfer function and use the mixer to compensate for any overly bright instrument sources. Above 95dBA (or so) I start to feel the need to apply additional shaping equalization to the baseline “flat” transfer function. If I don’t do this additional shaping equalization, I feel like I am using most of the the board eq to do the same types of equalization on most every channel.

Shaping equalization typically goes about like the Raphson-Dodson or ISO equal loudness contour curves, which have a broad smooth increase in sensitivity from about 1khz to 8khz. In general I start at 4khz, and spread out to each side. Above a certain HF cut level I find the octave centered around 400hz may need some minor tweaking to not be “honky”.

Above 8khz, i like to leave the sparkle. That can mean anything from a few dB shelving boost, to a few dB of cut depending on the room/system. Also, it seems that some systems have one particularly prominent tone in that top octave, so sometimes a single parametric cut up there is in order. Play some pink, sweep the top octave, and see if it clears out a “spike”.

Ideally the above equalization could be implemented with something like a BSS 901, so that the cuts were removed at moderate levels, retaining a “flat” curve at lower levels. I definitely think dynamic equalization has a place in system tuning for live sound applications.

----

Hopefully others will share their thoughts.

Jamie
August 11th, 2008, 10:08 PM
Damn! Welcome aboard!

Very well said.

Just to throw in a comment for the heck of it . . . after an initial pass at a tuning, I use the Spectrograph and listening (a few different male and female vocals, some piano, etc,) to quickly find the "spikes" and hollow notes. With a couple dB out with a couple filters I can clean up the Mids and Highs really quickly - faster than I can with straight measurement. The key, of course, is to get all of the heavey lifting done first.

Fantastic post! Keep 'em comin.

Zoinks,
-j

Joseph Pearce
August 12th, 2008, 09:15 AM
Great reading!

I'd like to mention that the old equal loudness curves (ISO 226) were updated in 2003- however I dont think there is information regarding levels at and above 100 phons...This may or may not be news to you and, in all honestly, I don't really know what to make of them...


re:

I definitely think dynamic equalization has a place in system tuning for live sound applications.

Aside from the BSS 901, what other tools have you used for this? And in what situations or program material? (ie. for use when going from speech level to band/playback)

Hope you dont mind a few follow-up questions. Thanks again for the great post!

phasetransitions
August 13th, 2008, 04:26 PM
Great reading!

I'd like to mention that the old equal loudness curves (ISO 226) were updated in 2003- however I dont think there is information regarding levels at and above 100 phons...This may or may not be news to you and, in all honestly, I don't really know what to make of them...


re:

I definitely think dynamic equalization has a place in system tuning for live sound applications.

Aside from the BSS 901, what other tools have you used for this? And in what situations or program material? (ie. for use when going from speech level to band/playback)

Hope you dont mind a few follow-up questions. Thanks again for the great post!

I was thinking of the latest 2k3 ISO curves when I wrote that. I don't really think I trust the average person's hearing accuracy above 100phons, but the chart gives enough of an idea of how things change to be useful.

Actually never used a 901 for this, but it seems the most logical piece to do this with. I have used an OmniDrive Compact + for this, and also fiddled around with the idea in Soundweb.

I should clarify that in a (green) Soundweb you are going to have build this by applying equalization to the sidechain of a compressor for the purpose.

Calvert Dayton
August 28th, 2008, 11:27 AM
Interesting topic, Phil. I wanted to add that I remembered reading that the X curve, or at least the fact of its persistence into modern cinema sound systems, had something to do with psychoacoustics -- i.e., that the perception of depth that you get when viewing a film on a really large screen comes with an unconscious expectation of some amount of roll-off on the high end in particular, to mimic the effects of air loss in the real world. This always seemed to make intuitive sense and I note that earlier room curve recommendations for cinema sound systems, before RTAs came along, were arrived at via listening tests and measurements at the power outputs of the amplifiers.

Ioan Allen mentioned psychoacoustics as a possible explanation for the perceived over-brightness of cinema sound tracks played back systems tuned flatter in large rooms in his 2006 paper (http://www.dolby.com/uploadedFiles/zz-_Shared_Assets/English_PDFs/Professional/Dolby_The%20X-Curve__SMPTE%20Journal.pdf) on the history and origins of the X curve, but he didn't explore it. Dunno if that meant he didn't buy it or just didn't want to go there. But I finally got around to tracking down the article wherein I had remembered reading that. Here's a link (http://www.micasamm.com/publications/surround_0100b.htm), FWIW.

I don't see any reason why both things couldn't be true though. Certainly if it were the case that visual cues created a subjective expectation of roll-off on the high end and conventional analysis methods tended to result in somewhat overly bright system tunings, the latter would only exacerbate the former unless compensated for somehow.

On a somewhat related note, since you brought it up, it's actually a well kept (OK, not all that well kept) secret that time window selections for Smaart originally had everything to do with performing the minimum possible number of the smallest possible FFTs, due to CPU budget constraints, and nothing whatsoever to do with much of anything else. It simply wasn't possible to do 32k FFTs in real time in a field measurement system 10 years ago, so if you wanted 3/4 second's worth of low end frequency response in a full-band FFT-based measurement, those were the hoops you had to jump through to get it.

About the only time I can recall the question of time windows ever coming up in the discussion as a driving design consideration early on was when we went from 16 to 24 points per octave in Smaart 3 and the same three FFTs we had used in version 2 no longer sufficed to get all three of the top three octaves. At 48k SR we were just point or two shy of 24 in the (~)4k octave band with the 256-point FFT we had used for the high end in version 2. But we ended up splurging on a fourth FFT just to pick up the the 4k band, rather than using a single 512-point FFT to grab all three top octaves, b/c it was felt that there was a benefit to keeping the time window shorter for the top two octaves to screen out short reflections.

This is not to say that observations regarding similarities between how FPPO measurements work and and how our ears work aren't perfectly valid, only that this was serendipitous and not a planned result. The insight followed the implementation rather than inspiring it.

phasetransitions
September 1st, 2008, 11:32 AM
Interesting topic, Phil. I wanted to add that I remembered reading that the X curve, or at least the fact of its persistence into modern cinema sound systems, had something to do with psychoacoustics -- i.e., that the perception of depth that you get when viewing a film on a really large screen comes with an unconscious expectation of some amount of roll-off on the high end in particular, to mimic the effects of air loss in the real world. This always seemed to make intuitive sense and I note that earlier room curve recommendations for cinema sound systems, before RTAs came along, were arrived at via listening tests and measurements at the power outputs of the amplifiers.

This is probable, too, I suppose. The only questioin I have with it is that the reference theaters that projects are mixed down in are balanced the same way, so in the end, it becomes and eq decision on the part of the engineer mixing the film. Maybe they found they were applying this basic eq shape to most sources, and just decided to move it down stream?



On a somewhat related note, since you brought it up, it's actually a well kept (OK, not all that well kept) secret that time window selections for Smaart originally had everything to do with performing the minimum possible number of the smallest possible FFTs, due to CPU budget constraints, and nothing whatsoever to do with much of anything else. It simply wasn't possible to do 32k FFTs in real time in a field measurement system 10 years ago, so if you wanted 3/4 second's worth of low end frequency response in a full-band FFT-based measurement, those were the hoops you had to jump through to get it.

That is new to me, but it make sense! Keep FPPO, its worth it :-)

I just downloaded the SMAART 6 demo. It seems very clean, and has ASIO drivers (best feature).

Two questions. What is the deal with the vector averaging? And do you have a suggested frame overlap in IR mode?