From: JasonCuadra@remove-this.-stec-power.com

Thanks a bundle for the transcript, Curtis. Thanks too to Bob.

It confirmed what I suspected is the right way: EQ the larger "repeat

offenders". This is the approach I did for my car, with good results, using a

quasi-parametric EQ. Interestingly I did my multi-location measurements in the

vicinity of the driver's head. Result was it sounded good from there, but when

you move your head to the right, between the driver and passenger, it sounds

bad. My only problem now is a kind of upper-mid harshness on some CDs which I

can't find in my EQ or distortion measurements.

The article leaves out some of the finer technical detail. It doesn't answer my

questions: What FFT time length do you use for the different frequency ranges?

Bob only suggests enough time length to get 1/24th octave resolution. What if

you only have a 1/3rd octave EQ? What smoothing do you use? What do you do if

you get different f/r curves with different time windows? Many references

suggest that the ear is sensitive ( in a perceived f/r sesnse) to roughly only

1/3rd freq bands. What is the perceived response if you have several very

narrow notches? Is 1/3rd octave smoothing accurate?

Regards,

Jason

From: JasonCuadra@remove-this.-stec-power.com

To: Lauduserslist 

Hi guys,

Bob wrote back and graciously answered my questions:

 ---------

Q:What FFT time length do you use for the different frequency ranges?

Bob only suggests enough time length to get 1/24th octave resolution.

A: You need 2.5 ms for the octave above 10kHz. Double that length for each each

successive octave as you go down. There are two manufacturers that already do

this for you

Meyer Sound's SIM and SIA's SMAART. 6o6

Q: What if you only have a 1/3rd octave EQ?

A: Chances of accurate filter placemant are

greatly reduced. Sometimes you can get lucky. Sometimes you can't 6o6>>

Q: What smoothing do you use?

A: You don't need smoothing. The best smoothing is done

> intellectually - by obswerving the full resolution data in multiple positions

and

> seeing the trends. It is an acquired skill.6o6 >>

Q: What do you do if you get different f/r curves with different time windows?

A: Use the octaves

> with the same resolution and tthrow out the rest - i.e. use the octave with

1/24th

> octave res and ignore the rest, then go to the next time window/octave.

Q: Many references suggest that the ear is sensitive ( in a perceived f/r

sesnse) to roughly only 1/3rd freq bands.

Very inconclusive- big diference in perception of peaks and

> dips. Take a tenth octave peak on you parametric and sweep it. You will have

no problem tracking it. Make it a dip and it is much harder. The most critical

wide peaks - the whalebacks. They can be only a few dB but if they are wide they

will really be audible. Those are my #1 tsargets 6o6

Q: What is the perceived response

> if you have several very

> narrow notches? Is 1/3rd octave smoothing accurate?

A: < Mathematically it can be.

> Operationally, no. It snowplows the trends. 6o6 >>

--------------

Hi Bob,

Thanks for your reply.

I hope you don't mind, I have a few more follow up questions.

>> You need 2.5 ms for the octave above 10kHz. Double that length for each each successive octave as you go down. There are two manufacturers that already do this for you Meyer Sound's SIM and SIA's SMAART. 6o6

Essentially you want 1/24th octave resolution for every octave band, right? Would it work to make a very long FFT, to make it 1/24th octave at say 100Hz Hz (250mS), but of course it would be much finer at higher octaves, then do 1/24th octave smoothing, and use that for EQ'ing?

Also, is there any spatial averaging technique that works, or do you eyeball this too? I tried the following experiment in my living room, to EQ 500 Hz up:

1) capture FFTs at 3 positions a foot apart sideways from listening position, ~12 mS long

2) Use Excel to (a) turn dB values from LAUD into power i.e. 10^(dB/10)

(b) average the power readings per frequency, from the 3 measurements.

3) turn it back into dB

Interestingly, the results converged into my 1/3rd octave smoothed results from the single central listening position. This experiment is why I just

And then, the above curve looked similar to a simple 1/3rd RTA of it! Dohh!

But then that was only 1 experiment (tedious so I only did it once). Have you heard of a similar experiment done more rigorously?

>> .. you said one sometimes gets lucky with 1/3rd octave EQ's.

In practice, in a typical room ( my interest is really audio reproduction, (home and car), not sound reinforcement setups ), how many parametric EQ bands does one need? (After some modest sound control steps in the home, can't do this in a car :-) )

>> You don't need smoothing. The best smoothing is done

> intellectually - by obswerving the full resolution data in multiple positions and seeing the trends. It is an acquired skill.6o6 >>

I suppose the natural thing to ask is, what's the procedure on "eyeball" smoothing? :-) You already mentioned that the whaleback peaks are the #1 targets. This was the gist of my question about what to do about "grass" or reflection induced multiple narrow dips w/in a single 1/3rd octave band.. Does the ear hear them? If I only have a 1/3rd octave EQ, do I bring up the average level of that band to make up for the narrow dips?

Jason

--------------

Q: Essentially you want 1/24th octave resolution for every octave band, right? Would it work to make a very long FFT, to make it 1/24th octave at say 100Hz Hz (250mS), but of course it would be much finer at higher octaves, then do 1/24th octave smoothing, and use that for EQ'ing?

A: No that is not a good plan. The smoothing can give you the constant resolution but the time window is too long in the high end. This makes the data inclousive of late reflections that are too impractical to EQ. It is MUCH better to keep the time windows short in the high end and expand them as you go down. This keeps a constant relationship between the period (1/F) of the frequencies involved to the time period of the window. Therefore a constant proportion of direct and early reflections are contained in the data.

6o6

Q: Also, is there any spatial averaging technique that works, or do you eyeball this too?

A: Spatial averaging is best done in my experience by comparing

> successive traces visually. A mathematical addition is not valid to to

> considerations regarding the coherence factor of each trace. 6o6>>

Q: I tried the following experiment in my living room, to EQ 500 Hz up:

>

> 1) capture FFTs at 3 positions a foot apart sideways from listening position,

> ~12 mS long

>

> 2) Use Excel to (a) turn dB values from LAUD into power i.e. 10^(dB/10)

> (b) average the power readings per frequency, from the 3 measurements.

>

> 3) turn it back into dB

>

> Interestingly, the results converged into my 1/3rd octave smoothed results from

> the single central listening position. This experiment is why I just

>

> And then, the above curve looked similar to a simple 1/3rd RTA of it! Dohh!

A: It will look similar to an RTA which is exactly why I recommend against it. In my experience the phase and coherence data are vital to the decision making process.

Q: In practice, in a typical room ( my interest is really audio reproduction, (home and car), not sound reinforcement setups ), how many parametric EQ bands does one need?

A: 5-6bands. Above 6 you are in need of a non-eq solution. 6o6 >>>

> (After some modest sound control steps in the home, can't do this in

> a car :-) )

Q: This was the gist of my question about what to do about "grass" or

> reflection induced multiple narrow dips w/in a single 1/3rd octave band.. Does the ear hear them?

A: Yes but not as peaks and dips. They are heard more as

> "grain" or "blur" 6o6>>>.

Q: If I only have a 1/3rd octave EQ, do I bring up the

> average level of that band to make up for the narrow dips?

A: <<<<Not recommended 6o6