Actual Double Blind Studies

[Jolly Joker]

It refers to the sound pressure at various angles to the speaker, and how that characteristic progresses through the frequency range. It’s pretty widely understood that gradual or smooth changes in directivity are preferred to abrupt changes, which cause incongruity between direct and reflected sound.

I know that much, I just don't see where there's discussion of it in this thread

 

[cavedriver]

Can you give me a pointer? The word "directivity" itself isn't mentioned that I can find

oops, I was thinking we were in another thread (the what's left to measure thread and the lack of high end speaker thread both get into this issue) since you were asking about DSP'ing speakers into equivalence. ahofer's reference is a good place to start, but the answer remains the same- DSP can't make two speakers sound the same in all respects and cannot correct for some differences in rooms.

 

[Jolly Joker]

oops, I was thinking we were in another thread (the what's left to measure thread and the lack of high end speaker thread both get into this issue) since you were asking about DSP'ing speakers into equivalence. ahofer's reference is a good place to start, but the answer remains the same- DSP can't make two speakers sound the same in all respects and cannot correct for some differences in rooms.

Aah, I see the misunderstanding. I wasn't talking about faking different speakers with DSP; I meant testing what differences people can hear and what sound they prefer using just DSP for the double blind test, using actual measurements from a microphone. Compare source material to what the mike picks up and ignore the fact that speakers even exist.

You do run into the psychoacoustic thing where what people expect to hear is based on the acoustics of the room.

 

[cavedriver]

Aah, I see the misunderstanding. I wasn't talking about faking different speakers with DSP; I meant testing what differences people can hear and what sound they prefer using just DSP for the double blind test, using actual measurements from a microphone. Compare source material to what the mike picks up and ignore the fact that speakers even exist.

You do run into the psychoacoustic thing where what people expect to hear is based on the acoustics of the room.

I'm not sure I'm understanding how you would use DSP in this testing, or what setup you are proposing. For testing the limits of human perception there have of course been many methods used, including digitally generated test signals. What do you mean by "actual measurements from a microphone" in this context? The mics would be picking up something, presumably the "source material", so what would the source material be compared to? other, different source material?

 

[Jolly Joker]

I'm not sure I'm understanding how you would use DSP in this testing, or what setup you are proposing. For testing the limits of human perception there have of course been many methods used, including digitally generated test signals. What do you mean by "actual measurements from a microphone" in this context? The mics would be picking up something, presumably the "source material", so what would the source material be compared to? other, different source material?

You would have, for example, a WAW file being played and a WAW file recorded at the listening position. Then you could vary the output being played in the room, affecting the subjective experience of the listener and the objective measurement of the microphone. The differences of the input signal and the measured output would be compared with the listener's responses. Do people prefer a 2 db boost below 400 Hz or not? Can they hear the difference?

 

[mtmpenn]

I’d be curious to know what a power and sample size calculation for the smallest meaningful difference looks like.

For those who are not familiar, the power of a study is defined as the probability of failing to identify a difference that is actually present.

In biomedical research large controlled trials are typically designed with 80 or 90% power.

Despite accepting a 10 or 20% chance that you fail to detect a true difference between groups, if the meaningful difference between groups is small, sample size gets big fast - often requiring thousands of people.

So, if we really wanted to know if speaker x was say 5% better than speaker y, how many people or blinded trials for a single individual would be necessary to confidently conclude a difference?

I don’t have the stats skills to figure this out but I bet it’s more than most would think.

 

[Dave Bullet]

I've used ABX for active (digital) crossover experimentation. The beauty with a software DSP is you can script it to flick between 2 possible crossover topologies, and select these randomly playing them. That way I have no idea which one the computer is playing. i.e. I give it crossover 1 and 2 on the command line. It randomises these as "A" and "B" then randomly plays as many samples of a recurring or continuing track, pausing inbetween samples for me to record whether I am hearing "A" or "B". The script records the evidence I can check afterwards. I've used this a few times to see if I could consistently tell one from another (in order to have confidence in any preference).

the same could be done if you had enough downstream components (DAC, amps) to ABX 2 pairs of speakers. Assuming you SPL matched them and could position them as close as possible so that position change wasn't detectable.

 

[cavedriver]

You would have, for example, a WAW file being played and a WAW file recorded at the listening position. Then you could vary the output being played in the room, affecting the subjective experience of the listener and the objective measurement of the microphone. The differences of the input signal and the measured output would be compared with the listener's responses. Do people prefer a 2 db boost below 400 Hz or not? Can they hear the difference?

I'm not sure what you would measure there that hasn't been studied in a number of other studies. The limits of human hearing are apparently well documented. As far preferences, I think that's covered in the research done at Harman and frequently referenced on this site. I haven't read it myself but a large number of the people here are well acquainted with it. Basically it says that the human preference (and I don't remember the margin) is for a flat response.