Re: Filter design procedure: treble and bass part 2

[Mitch Global]

Don’t know how to attach to an existing post…

 

Hey Davide, sorry for not getting back to you sooner re: linear vs minimum phase targets. I can’t remember, but I have been using minimum phase targets for both frequency and TTD as my default config.

 

With respect to the turning off prevent treble/bass boost, attached are a few files to compare.

 

"Default freq corr.jpg" uses the Audiolense out of the box frequency correction preset. It has the prevent treble/bass boost checked on and default 6 db correction. Btw, I am still using the B&K house curve as the reference target, attached as "bandk.jpg". The way I interpret the B&K curve is -.5 db at 200 Hz, -3 db at 2 KHz, and -6 db at 20 Khz.

 

Back to the default freq corr, If you look closely at the frequencies extremes, in the low end, one speaker is rolling off at 30 Hz. and the other 40 Hz. At the top end, it rolls off steeply around 17.5 KHz.

 

I unchecked the prevent treble/bass boost and also played with the amount of db correction. I ended up with 18 db of correction after listening to 12 and 15 See "Corr Proc Designer.jpg".

 

Have a look at "18 db corr freq response.jpg". Now both speakers go to the low 20’s and I roll off below that, in the target, to keep the very low end tight. The top end goes out to 23.5 Khz.

 

Attached is the target I used, based on the B&K curve "B&K Minimum Phase adj4.jpg". It is almost identical to the B&K curve except with a drop of -1 to -2 db from the reference at 20 KHz. It may sound just a bit laid back at the very top. If so, just add back in a db or 2 to bring it back in line with the B&K house curve – see "B and K target adj.jpg". 

 

Play around with the frequency extremes to tailor fit your requirements. Note that a small tilt adjustments (1 or 2 db) make big timbre differences.  

 

I am still fine tuning myself, but compared to the frequency limits I had before, this a another big improvement. Aside from the +- 3db variance at 200 Hz (and ripple from just over 100 hz to 550 Hz), the variance across the band is +- 1db. Sounds like a fine tuned instrument .

 

Hope that helps.

 

Hi Bernt, any chance that those prevent treble/bass boost checkboxes could turn into adjustable high and low pass filters?  Or is the idea that you draw in the roll offs in the target?

 

Happy Holidays!

 

Mitch

[lasker 98]

Hi Mitch,

I've also been experimenting with prevent treble/bass boost settings.
My current best result is to have the target curve follow the
measured, smoothed frequency response on the bass end, with no rolloff
added at the high end. I have both prevent treble and bass boost
unchecked, as well as partial correction selected. This is then set
for both right and left speakers, with no correction above 20,000 Hz.
The other settings in that window all at 0.

This gives me a simulated result curve that follows the target
perfectly in the bass (down to about 10 Hz, below that it pretty much
follows the measured response) and at the high end it follows the
target perfectly out to 20000 Hz and above that it follows the
measured response perfectly. All in all, a very nice result.

I too am using the B&K target thanks to you. I originally was using
the target file you posted in an earlier post here. I then tweaked it
by eyeballing the actual picture of the curve you posted. Since then
I've found this through this link to DRC documentation:

http://drc-fir.sourceforge.net/doc/drc.html

"The bk.txt file follows the Bruel & Kjaer (i.e. Mœller)
recommendations for listening room frequency response, i.e. linear
from 20 Hz to 400 Hz, followed by a slow decrease of 1 dB per octave
up to 20 kHz." (you'll find that text with a graphic of target curves
about 1/4 of the way down that posted link page).

I've discovered that in the Audiolense Target Designer, if you right
click in the main window, a chart editor window pops up. Select data,
then you can enter the exact data points for the target. Not sure if
this is how everyone has always done it but I just recently discovered
it so thought I would pass it on just in case it's new to some.

For my target, the data points I'm using are 1) X=20, Y=-16, 2) X=28,
Y=0, 3) X=35, Y=0, 4) X=400 Y=0, 5) X=20000, Y=-6

Points 1, 2 and 3 give me the bass curve to follow the measured
response of my speakers, with an exact B&K target otherwise.

I'd be very interested in hearing thoughts or comments on this
approach.

One other question for you Mitch. I noticed in one of the images you
posted somewhere earlier, you show the J River convolver window. It
looks like you have loaded the .wav file for the correction? Based on
something I read in Audiolense help, I've been using the .cfg
correction file from the "CorrectionFiles" folder. is there a reason
for your choice or is it just how you've been doing it? One thing I've
found with the .cfg file, I listen to a lot of random playlists and
some songs will be 16-44.1, some 24-96 and J River when using the .cfg
file will switch seamlessly between formats. I'm assuming it's using
the correction filter on both formats. I'm not sure if it works the
same way using the .wav file. I'd be interested in hearing Bernt's
comment on this as well.

One last thing I just thought of as well Mitch. I notice you seem to
have done your measurements at 96000 Hz sample rate? I have no idea on
the pros and cons of what rate to choose for the measurement. I've
always just used the default 48000 Hz for mine. Again, I'd like to
hear what Bernt has to say about this as well.

Bill

>  Files to compare.zip
> 1587KViewDownload

[Mitch]

Hi Bill,

Excellent tips, thanks! I went back to my target and readjusted based
on right clicking and editing the data points, back to the original
B&K curve. As it turns out, because I did not have partial correction
turned on, I was getting excessive hi-freq boost that made me tweak
down the B&K curve at 20 KHz. Great tip to check partial boost on and
limit correction after 20 KHz.

re: http://drc-fir.sourceforge.net/doc/drc.html I had seen that
before, but thought it was identical to the one I had posted. Will
look closer.

re: .cfg file. What version of JRiver are you using? Just checking,
do you have resample all sources turned on? You have no resampling
right? I ask as I have not had too much luck with .cfg files.
According to: http://convolver.sourceforge.net/config.html you need
to create a filter list. I tried that and could not get it to
work...

re: measurement sample rates. I just chose 96Khz as that is what I
tend to listen at. I don't think it makes any diff.

Thanks again for the tips Bill.

Cheers, Mitch

> > 1587KViewDownload- Hide quoted text -
>
> - Show quoted text -

[Davide Hong Kong]

Hi Mitch,

Thanks for the post and the attachments, as always very interesting.

I have switched from linear to minimum phase filters and I felt a bit
more "cleanness" and sharper transients.

Regarding the correction, I have a big dip at 60hz and a big boost at
21 hz and have to deal with those, so correction I allow is a
considerable 25db.

My only complain is that it feels like dynamics are diminished... But
I am coming to realize there is not much dynamic to be had from cone
speakers once you take away room gain...

I continue of course to use the B&K target (although I roll off less
in the treble as I listen far from the speakers).

One question for you guys: when you look at your simulated (corrected)
response around the target, which is the max deviation you get from
target? The simulated response I get is within 3.5 db of the target.

I am curious to know if you get a tighter response.

Interesting comment on the config files: I could not get any sound to
come out from J River (I am on version 16) using the .cfg files and
use the WAV version of the correction filter. But can only play a
certain sample frequency (48k) and use in J River output options
resampling all @ 48k and 32 bit output is mandatory to get sound with
filters saved at 48k.

Would you be so kind Bill to share the settings you have in convolver
to play the .cfg? Not using resampling in J River could result in a
sound improvement..

Wishing you all a merry xmas and happy new year.

Cheers
Davide

[lasker 98]

Hi,

I too have a big dip in response around 65 Hz. On my filtered
measurement it's down 25 db at 65, goes up to about -15 db, then back
down to -25 db around 90 Hz, then back up to between -10/-15 db, then
back to -25 db around 130 Hz.

My room is about 12' x 35' x 8', combined family room/kitchen. The
speakers are setup in the family room across the 12' wall. The
measurement above is taken about 17' away from the speakers, where I
sit facing across the kitchen table towards the speakers.

For that correction filter, I'm able to get away with using a max
boost of 15 db and my simulated result is down less then 3 db at the
frequencies I gave above. No Idea how I'm correcting a 25 db dip to
less than a 3 db dip with only 15 db of boost but that's what I get.
If I increase the max boost I can improve even more on the final 3 db
dip but something I've found is I pay very close attention to the
maximum amplitude of the simulated result on the Impulse Response
graph. To me, the correction filters with the highest amplitude
impulse response (remember, this is the corrected, simulated result
I'm talking about) "sound" the best. Hard to describe exactly, but I'm
thinking it's because of increased dynamics? Even if the frequency
response has more +/- db variation in it, the filter with the highest
impulse response amplitude sounds better, at least to me. So I'm
willing to accept a slightly worse looking simulated result for a
trade off of a greater impulse response amplitude. I'd really be
interested in hearing feedback from others on this. This whole process
is new to me so I have no idea if what I'm looking at is correct.

On the big dip around 65 Hz, for my more serious listening, I sit in
the family room about 8' from the speakers. This filtered measurement
shows a single dip of about -32 db around 65 Hz. For this correction I
have to use max boost of 25 db. This gives about the same -3 db dip in
the corrected simulated response.

For the config files, I've done nothing with my convolver other than
install it (I wouldn't know what to do with it anyways). I'm using
JR17 and have been since I started with Audiolense. The only
resampling I have set in J River is "less than 44,100 Hz to 44,100 Hz"
and everything from 176,400 Hz and above resample to 96,000 Hz (my dac
is max of 96k Hz).

I'm using the 2.0 44.1 .cfg file from my Audiolense 4.2/
CorrectionFiles folder. Also, when I save my filters, I check the
boxes for 44.1, 48 and 96, which creates 3 separate files, No idea if
this is how I'm supposed to do it but that's what I've done from the
beginning. If I remember right, I think I initially tried using
the .cfg file from the Audiolense/Correction folder and I don't think
that worked. This was based on reading the help file for Audiolense.
The specific section that refers to this is "Different playback
methods". Reading that I took it to mean use the .cfg file from
Audiolense/Correction folder but as I said, that didn't work for me.
If you guys are having problem, maybe try using the .cfg file from the
Audiolense/CorrectionFiles folder.

One other thing with convolver setting. Once the .cfg file is loaded,
at the bottom of the window it gives estimated gain and peak gain. I
always set the attenuation slider at the inverse of what the peak gain
is. For example my current .cfg file peak gain shows 6.9 db, so I set
attenuation slider at -6.9 db. I read about this on the convolver/
sourceforge site.

I hope to hear back with any comments or suggestions. Like I said
before, this is all new to me. I have no idea if what I'm doing is
correct or not.

Bill

> > > - Show quoted text -- Hide quoted text -

[Alan Jordan]

Hi,

Am I incorrect in thinking its a bad idea to try to fix a deep room null by allowing a great amount EQ boost?  You can EQ a room null all you want but it is still going to be a null, but at the same time you are making your amps and speakers work much harder and probably causing more distortion in the end result.  Your brain does a great job filling in narrow room nulls for you.  I thought that, taking psycho-acoustics into account, it was sufficient to correct to your chosen target curve at a third of an octave resolution. Please correct me if I am mistaken

I think it shows you are correcting a 25 dB dip with 15 dB of gain because the simulated result is probably graphed at a lower octave resolution than the measurement, but I am only guessing.

Alan

--
You received this message because you are subscribed to the Google Groups "Audiolense User Forum" group.
To post to this group, send email to audio...@googlegroups.com.
To unsubscribe from this group, send email to audiolense+...@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/audiolense?hl=en.

--
http://www.alanjordan.org

[Alan Jordan]

One point regarding the gain values of a filter.  The peak gain value would be if you were playing a broadband source like pink noise at 0 dB.  Its imagine that setting your attenuation slider at -6.9 dB will cause an unnecessary reduction in resolution because your estimated gain is already at -6.9 dB.  If you mostly listen to highly compressed pop music, then you will probably need lower gain on the slider than if you listened to mostly classical and jazz.  With my filters, I show an estimated gain of -11 dB, and a peak gain of 11 dB, but I keep my slider at +4dB and have not clipped at all at that setting.

If your DAC does not have meters that show the input level, in JRiver you can see what the peak level by opening the DSP Studio and looking in the lower left corner while music is playing.  As long as the level stays below 100 percent, you are good.

Alan

On Wed, Dec 21, 2011 at 9:14 AM, lasker 98 <stree...@gmail.com> wrote:

One other thing with convolver setting. Once the .cfg file is loaded,
at the bottom of the window it gives estimated gain and peak gain. I
always set the attenuation slider at the inverse of what the peak gain
is. For example my current .cfg file peak gain shows 6.9 db, so I set
attenuation slider at -6.9 db. I read about this on the convolver/
sourceforge site.

I hope to hear back with any comments or suggestions. Like I said
before, this is all new to me. I have no idea if what I'm doing is
correct or not.

--
http://www.alanjordan.org

[Mitch]

Hi Davide,

re: One question for you guys: when you look at your simulated

(corrected)
response around the target, which is the max deviation you get from
target? The simulated response I get is within 3.5 db of the target.

Here is my uncorrected freq response of my speakers/room:

http://i1217.photobucket.com/albums/dd381/mitchatola/uncorrected.jpg

And corrected with target:

http://i1217.photobucket.com/albums/dd381/mitchatola/correctedplustarget.jpg

At 200 Hz, my max deviation is +- 3.5db of the target, but on either
ends of that, I can get +- 1db of the target. Btw, it still blows my
mind looking at the difference between the uncorrected and corrected
frequency response. I tired to keep the vertical and horizontal
scales the same to really bring home how well this software works.
Amazing!

Hi Bill, it would be cool to see your impulse responses.... I am
using JRiver 17 too and still can't get convolverVST to auto switch
like you are able to get. I exchanged emails with the author of
convolverVST about auto switching files. There is a way, but I was
not able to figure it out. Maybe others can. I will follow up once I
find the emails as it would be nice not to have to either resample
everything or manually switch...

Hi Alan, yes, my understanding is that we do not hear differences less
than a 1/3 of an octave. However, somewhere I read, it was closer
between 1/3 and 1/6 octave: http://www.silcom.com/~aludwig/EARS.htm
I believe that Audiolense displays 1/12 octave in it's graphs. I only
say that as I used REW to correlate what I was measuring in Audiolense
and superimposed the 2 graphs here: http://i1217.photobucket.com/albums/dd381/mitchatola/superimposed.jpg
and REW was set for 1/12 octave smoothing.

With respect to gain adjustments and peak levels. I use the TT
Dynamic Range Meter plugin that you can get here: http://dr.loudness-war.info/
if you look under the software links on the top right, you will see
the download for Windows. This is what the meter looks like: plugged
in JRiver: http://i1217.photobucket.com/albums/dd381/mitchatola/TTDRMeter.jpg

On my filters, I have an estimated gain of -18 db and a peak gain of
4.2 db. So I have cranked up the slider to about +11 db and watch the
dynamic range meter for any signs of clipping.

Happy Holiday!

Mitch

On Dec 21, 9:59 am, Alan Jordan <aljor...@gmail.com> wrote:
> One point regarding the gain values of a filter.  The peak gain value would
> be if you were playing a broadband source like pink noise at 0 dB.  Its
> imagine that setting your attenuation slider at -6.9 dB will cause an
> unnecessary reduction in resolution because your estimated gain is already
> at -6.9 dB.  If you mostly listen to highly compressed pop music, then you
> will probably need lower gain on the slider than if you listened to mostly
> classical and jazz.  With my filters, I show an estimated gain of -11 dB,
> and a peak gain of 11 dB, but I keep my slider at +4dB and have not clipped
> at all at that setting.
>
> If your DAC does not have meters that show the input level, in JRiver you
> can see what the peak level by opening the DSP Studio and looking in the
> lower left corner while music is playing.  As long as the level stays below
> 100 percent, you are good.
>
> Alan
>

[lasker 98]

Hi,

I discovered last night that I'm wrong about the automatically
switching of the sample rates in convolver. I found some of the tracks
I thought were 96khz were mis-tagged. I tried playing some known 96
khz files and they do not play unless everthing is upsampled to 96
khz. I'm very sorry for the confusion. I'm sure at this time of year
no one wants to be wasting time, especially because of wrong
information.

I'm also having computer problems. I may end up having to re-format. I
don't think I'll be having much more time to play with this over the
holidays.

Thanks Alan and Mitch for the input on monitoring the playback levels.
I'll look into that more for sure as well as your other comments Alan.

Again, I'm very sorry for leading you astray on the sampling
switching. It would be great to have it work but at this point mine
seems to be working as described by others here.

I hope everyone enjoys their holidays.

Bill

On Dec 21, 5:14 pm, Mitch <mitchglo...@gmail.com> wrote:
> Hi Davide,
>
> re: One question for you guys: when you look at your simulated
> (corrected)
> response around the target, which is the max deviation you get from
> target? The simulated response I get is within 3.5 db of the target.
>
> Here is my uncorrected freq response of my speakers/room:
>
> http://i1217.photobucket.com/albums/dd381/mitchatola/uncorrected.jpg
>
> And corrected with target:
>

> http://i1217.photobucket.com/albums/dd381/mitchatola/correctedplustar...

> > --http://www.alanjordan.org- Hide quoted text -

[Mitch]

Bill, no worries man, all good. I hope you don't have to re-format.

It seems there is a way for convolver to automatically switch sample
rates on the fly.

If you go to http://convolver.sourceforge.net/config.html and scroll
down to the section called Filter List, there is a description on how
to automatically switch between filters.

I tried a few variations of that and could not get it to work. So I
contacted John Pavel the author of convolverVST. He suggested that it
is more likely to be an end of line problem (some files use return
+newline and some newline only (which I suspect is the preferred
format). You could also try to fiddle with the very last character in
the file.

I have not had a chance to revisit this. Maybe over the XMas break...
or maybe someone else will have better luck than me...

In the meantime, http://www.youtube.com/watch?v=eTCbiJQXIMY&feature=share

Mitch

On Dec 22, 4:04 am, lasker 98 <streetor...@gmail.com> wrote:

> > > --http://www.alanjordan.org-Hide quoted text -
>
> > - Show quoted text -- Hide quoted text -

[Bernt Rønningsbakk]

As for measurement I recommend 48 kHz for mid fi sound cards as some of
those have their best performance there. Some class D amplifiers will have
high frequency noise and then 48 can be the best as well.

But if you have none of those issues 88 or 96kHz may be the best way to go,
just for the added bandwith in the high treble. But unless the system has
issues, the choise of measurement FS is nitpicking and I wouldn't bet on it
making an audible difference.

Kind regards,

Bernt

[Bernt Rønningsbakk]

The time domain is important with regards to how much dips can and should be corrected. Quite often, actually, a lot can be done inside a short time window. Other dips requires too much time to be effectively corrected.

 

As long as sober values are used for TTD window and correction window, Audiolense will make a controlled effort to correct a dip. And therefore there will often be residuals of dips after correction even though the residual is inside the scope of the max correction boost. In practice, the max correction boost will only make a difference when a lot of correction can be done with a limited effort.

 

 

Kind regards,

 

Bernt

[Bernt Rønningsbakk]

Hey Mitch,

 

The idea is to draw the roll offs in the target.

 

 

Kind regards,

 

Bernt

 

--