Achieving 'equal loudness' for A/B testing

I'm trying to A/B test a filter, and I'd like to control for the human
preference for louder signals. Is there a standard way to do this?
Right now I'm normalizing to the avg RMS level of the a-weighted
signal. I'm hoping to achieve two signals with the same subjective
loudness so there's no 'louder signal' bias.

I'm a-weighting the signal before measuring the RMS in an attempt to
compensate for reduced sensitivity to low/high frequencies (as the
filter dramatically alters the spectral content it seemed like
normalizing to the straight RMS might still result in a different
subjective loudness... e.g. if the two signals had the same avg power,
but in one the energy went into 1khz and in the other the energy went
into 40Hz?).

Does this approach make sense to other people? Is there a better one?

thanks,

-Seth

On Thu, 27 May 2010 12:04:04 -0700, snickell wrote:
[a-weighting RMS power to make equal loudness]
> Is there a better one?

Officially, ITU-R BS.1770 is supposed to be a better measure of loudness 
than A-weighting.

There are some links to useful papers in this discussion:
http://www.hydrogenaudio.org/forums/lofiversion/index.php/t47154.html

Cheers,

-- 
Andrew

Andrew Reilly <areilly---@bigpond.net.au> wrote:
> On Thu, 27 May 2010 12:04:04 -0700, snickell wrote:
> [a-weighting RMS power to make equal loudness]
>> Is there a better one?
 
> Officially, ITU-R BS.1770 is supposed to be a better measure of loudness 
> than A-weighting.
 
> There are some links to useful papers in this discussion:
> http://www.hydrogenaudio.org/forums/lofiversion/index.php/t47154.html

The wikipedia page tell some of the problems with A weighting,
though not so much for solutions.

I was just wondering, if you take two similar but different audio 
signals and switch between them at, say, 1kHz, how the sound
changes with relative amplitude.  Assuming no definite phase
relationship between the two.

-- glen

>Andrew Reilly <areilly---@bigpond.net.au> wrote:
>> On Thu, 27 May 2010 12:04:04 -0700, snickell wrote:
>> [a-weighting RMS power to make equal loudness]
>>> Is there a better one?
> 
>> Officially, ITU-R BS.1770 is supposed to be a better measure of loudness

>> than A-weighting.
> 
>> There are some links to useful papers in this discussion:
>> http://www.hydrogenaudio.org/forums/lofiversion/index.php/t47154.html
>
>The wikipedia page tell some of the problems with A weighting,
>though not so much for solutions.
>
>I was just wondering, if you take two similar but different audio 
>signals and switch between them at, say, 1kHz, how the sound
>changes with relative amplitude.  Assuming no definite phase
>relationship between the two.

None of these weighting schemes capture much of the perceptual element.
That factor which makes us want to turn up the radio when a great piece of
music comes on, because it sounds so much quieter than the Barry Manilow
song which preceded it. Your 1kHz chopped signals will probably sound
horrible, and are therefore likely to be perceived as loud.

Steve

steveu <steveu@n_o_s_p_a_m.coppice.org> wrote:
(snip, I wrote)

>>I was just wondering, if you take two similar but different audio 
>>signals and switch between them at, say, 1kHz, how the sound
>>changes with relative amplitude.  Assuming no definite phase
>>relationship between the two.
 
> None of these weighting schemes capture much of the perceptual element.
> That factor which makes us want to turn up the radio when a great piece of
> music comes on, because it sounds so much quieter than the Barry Manilow
> song which preceded it. Your 1kHz chopped signals will probably sound
> horrible, and are therefore likely to be perceived as loud.

The idea, though I don't know that it actually works, was that
as the amplitudes got closer the amplitude of the chopped signal
would get smaller.  Even more, but less obvious, than it 
would follow the appropriate weighting.  

-- glen

On Thu, 27 May 2010 12:04:04 -0700 (PDT), snickell
<snickell@gmail.com> wrote:

>I'm trying to A/B test a filter, and I'd like to control for the human
>preference for louder signals. Is there a standard way to do this?
>Right now I'm normalizing to the avg RMS level of the a-weighted
>signal. I'm hoping to achieve two signals with the same subjective
>loudness so there's no 'louder signal' bias.
>
>I'm a-weighting the signal before measuring the RMS in an attempt to
>compensate for reduced sensitivity to low/high frequencies (as the
>filter dramatically alters the spectral content it seemed like
>normalizing to the straight RMS might still result in a different
>subjective loudness... e.g. if the two signals had the same avg power,
>but in one the energy went into 1khz and in the other the energy went
>into 40Hz?).
>
>Does this approach make sense to other people? Is there a better one?

Maybe you could conduct a pre-test with the subjects adjusting the
levels through paths A and B for equal subjective loudness (with
random order of presentation). Record the unweighted RMS levels
present after the adustment for equal loudness, and use those levels
during the test for whatever it is you are looking for in the main
test.
-- 
John

snickell <snickell@gmail.com> wrote in news:9c38ac81-c056-4f7d-94ed-
d607291d0304@s4g2000prh.googlegroups.com:

> I'm trying to A/B test a filter, and I'd like to control for the human
> preference for louder signals. Is there a standard way to do this?
> Right now I'm normalizing to the avg RMS level of the a-weighted
> signal. I'm hoping to achieve two signals with the same subjective
> loudness so there's no 'louder signal' bias.
> 
> I'm a-weighting the signal before measuring the RMS in an attempt to
> compensate for reduced sensitivity to low/high frequencies (as the
> filter dramatically alters the spectral content it seemed like
> normalizing to the straight RMS might still result in a different
> subjective loudness... e.g. if the two signals had the same avg power,
> but in one the energy went into 1khz and in the other the energy went
> into 40Hz?).
> 
> Does this approach make sense to other people? Is there a better one?
> 
> thanks,
> 
> -Seth

One thing to consider is that A-weighting is an approximation to NC (noise 
control) curves at about 65dB if I remember correctly (check this value) 

Hearing is actually somewhat flatter at higher levels. You can look at NC 
curves to see what I mean.

If you are normalizing levels of music, I expect these levels would be at 
somewhat higher SPL.

Al Clark
www.danvillesignal.com

On May 28, 9:45&#4294967295;am, Al Clark <acl...@danvillesignal.com> wrote:
> snickell <snick...@gmail.com> wrote in news:9c38ac81-c056-4f7d-94ed-
> d607291d0...@s4g2000prh.googlegroups.com:
>
>
>
>
>
> > I'm trying to A/B test a filter, and I'd like to control for the human
> > preference for louder signals. Is there a standard way to do this?
> > Right now I'm normalizing to the avg RMS level of the a-weighted
> > signal. I'm hoping to achieve two signals with the same subjective
> > loudness so there's no 'louder signal' bias.
>
> > I'm a-weighting the signal before measuring the RMS in an attempt to
> > compensate for reduced sensitivity to low/high frequencies (as the
> > filter dramatically alters the spectral content it seemed like
> > normalizing to the straight RMS might still result in a different
> > subjective loudness... e.g. if the two signals had the same avg power,
> > but in one the energy went into 1khz and in the other the energy went
> > into 40Hz?).
>
> > Does this approach make sense to other people? Is there a better one?
>
> > thanks,
>
> > -Seth
>
> One thing to consider is that A-weighting is an approximation to NC (noise
> control) curves at about 65dB if I remember correctly (check this value)
>
> Hearing is actually somewhat flatter at higher levels. You can look at NC
> curves to see what I mean.
>
> If you are normalizing levels of music, I expect these levels would be at
> somewhat higher SPL.
>
> Al Clarkwww.danvillesignal.com- Hide quoted text -
>
> - Show quoted text -

You can always turn it up until the ears start to bleed. This removes
the psychological effect from the measurement ;-)

Clay

On May 28, 10:05&#4294967295;am, Clay <c...@claysturner.com> wrote:
> On May 28, 9:45&#4294967295;am, Al Clark <acl...@danvillesignal.com> wrote:
>
>
>
>
>
> > snickell <snick...@gmail.com> wrote in news:9c38ac81-c056-4f7d-94ed-
> > d607291d0...@s4g2000prh.googlegroups.com:
>
> > > I'm trying to A/B test a filter, and I'd like to control for the human
> > > preference for louder signals. Is there a standard way to do this?
> > > Right now I'm normalizing to the avg RMS level of the a-weighted
> > > signal. I'm hoping to achieve two signals with the same subjective
> > > loudness so there's no 'louder signal' bias.
>

you can use a 1 kHz tone ( something midband of your filter) and set
the gain of your two channels ..

then apply the program material ...

if you MUST compare the loudness of program material directly (which
will be difficult to get very accurate) you should start here:

http://www.dorrough.com/

Mark

snickell <snickell@gmail.com> writes:

> I'm trying to A/B test a filter, and I'd like to control for the human
> preference for louder signals. Is there a standard way to do this?
> Right now I'm normalizing to the avg RMS level of the a-weighted
> signal. I'm hoping to achieve two signals with the same subjective
> loudness so there's no 'louder signal' bias.
>
> I'm a-weighting the signal before measuring the RMS in an attempt to
> compensate for reduced sensitivity to low/high frequencies (as the
> filter dramatically alters the spectral content it seemed like
> normalizing to the straight RMS might still result in a different
> subjective loudness... e.g. if the two signals had the same avg power,
> but in one the energy went into 1khz and in the other the energy went
> into 40Hz?).
>
> Does this approach make sense to other people? Is there a better one?

Hi Seth,

I would run each signal through a perceptual weighting filter and
compare the power of each output. I'm not sure where you can find
one - email me if you have trouble.

--Randy


-- 
Randy Yates                      % "And all you had to say
Digital Signal Labs              %  was that you were 
mailto://yates@ieee.org          %  gonna stay."
http://www.digitalsignallabs.com % Getting To The Point', *Balance of Power*, ELO