Apologies for not having a clue what forum to post this is in.

I'm in the process of writing some software to test the frequency response of my headphones using test tones, in response to what I found out in this thread.

My idea is to have the software play two tones of different pitches, one after the other. The user then has to respond with:
1) First tone is louder
2) Second tone is louder
3) Both tones are equal volume
4) No idea (probably not audible)
5) Play the tones again

When the user reports that one tone is louder than the other, the software will adjust the amplitude of one of the tones, making a guess such that the tones then appear to have the same amplitude.

The software will continue to play pairs of tones until it always gets "both tones are equal volume" at which point it will stop the test and save the results (possibly even make a graph..).

So....

Is this a good idea or is there a better way?

How many discrete frequencies should be used? I'm assuming the magical 12 per octave which gives 122 tones if we cover 20-20KHz.
I'm aware this may make the test quite long, so I could add a "save state" option so that the test wouldn't have to be completed in one session.

Should the tones played always be adjacent pairs, a set distance apart or a variable distance?
If the latter, any guidelines?
I think some distance should be used to prevent "rounding errors" accumulating.

Would it be enough to simply choose one (or a few) frequencies and compare this with all others?
This would reduce the complexity and duration of the test somewhat.

Is it better to randomly move around the frequency spectrum or for example, start with low frequencies and systematically work up to the higher frequencies?

How should the software guess the difference in amplitude between the tones? e.g. keep adding/removing 6db until the user response is inverted then remove/add 3db etc?

When does the test stop? When every frequency has been reported equal to some other frequency once? Twice?

I did some experimenting with this a while back and got some results out of it. I'm not really sure how useful the results were though.

First of all, you're not really testing the frequency response, you're testing equal loudness. Your results will need to be equalized again, using a Fletcher-Munson curve. I'd expect that curve to be slightly different for each person, although there's no good way for you to test it yourself without an already-equalized playback system.

I think merely having a volume control for one of the tones in each test is sufficient to get a good relative loudness difference. This doesn't need to be blind or anything.

I think a lot of the questions you're asking haven't really been answered, although I don't have a psychoacoustics text to see what professionals say. I'd try an all-combinations test from 100-20khz with 1.5 frequencies tested per octave - that's around 5! tests or 120 tests. Each test should take no more than 30 seconds, so you're only looking at an hour long test which shouldn't be toooo bad. Then plot the results as different data lines on an XY graph (one line for each frequency that is relatively compared with the otehr frequencies). There's going to be a spread of relative levels for each X frequency value, and the variance of that spread is going to tell you how necessary it is to use adjacent frequency points to do the comparisons.

Thanks - I thought this was going to be more complicated than it appeared to be!

> a Fletcher-Munson curve

I presume these relate to the ear (and maybe even the head) as a whole, so they may not match exactly to ear canal headphones? I'll do some more investigating to see how they're created.

My main concern is locating the sharp peaks, so maybe a manually-sweepable tone generator would be more useful, or a combination of the two.

Your goal is futile. There are no two ears with same response. You will be measuring your own and only your ears.
There is a special medical test procedure you ought to take time to time - timpanogram or audiometry. Unless you calibrate your setup precisely based on these results, you'll be "measuring" parrots.

http://www.audiometer.co.uk/
http://www.npl.co.uk/acoustics/techguides/audiometry/
http://www.speech.psychol.ucl.ac.uk/lectur...26/hearing2.pdf

I have some headphones I really like, apart from their perceived frequency response. I don't have this problem with my Sennheisers and other headphones (but I prefer the Etys, once eq'd), therefore I can perform the test with these also and compare the results. My HD-600s will be a good reference I think as they don't suffer from problems such as canal resonance.


Thanks for the info & links - I'll have a go with audiometer.