I gathered up a corpus of Head-Fi user reviews (the non-forum ones) of five different headphones: Audeze LCD-2, AKG K 701, AT M50S, Senn HD 598, and Senn HD 650; and harvested some frequency response data for those phones from Tyll's measurements at InnerFidelity. The aim was to map out some of the existence or not of correlation between a headphone's frequency response and the subjective reports of its acoustic properties using a corpus-based approach.
The analysis was done for a paper for course credit in the field of corpus linguistics. I'm not sure what my uni's stance is on students posting bits of papers not yet handed in, so I'll keep this cursory enough. I'm not sure this line of study entirely qualifies as a linguistic one, either, but luckily that's not much of a concern here.
In any case, main aspects of data, ranked descending by third column:
% of reviews with
vs 1 kHz level
The corpus I built included 79 reviews for the M50S while the other phones had about 40% less available reviews going for them. These 'missing' reviews were filled in via curve-fitting in SPSS - but even without fitting, by analyzing a strict number of the first 40 reviews for each phone, the overall results shown above held.
The third column is the percentage of reviews for that phone containing the word "accurate" - or "accuracy"/"accurately" - in reference to the phone (whether describing the mids, the treble, the soundstage, etc.). A second keyword, "transparent" along with its derivatives, was also used, but it failed to show up at all for three of the five phones, so it's ignored here.
The fourth column gives a number estimating the amount of energy arriving at the listener's eardrum along the mid frequencies of 2–4 kHz. The number was derived by eye from Tyll's PDF graphs by mentally averaging the grey (raw) lines at each of the three kHz markers in relation to the 1 kHz level and then summing the three averaged numbers together. Along with the mids, the bass and treble frequencies were evaluated in the same manner (40+100+200 Hz, and 8+9+10 kHz).
Like the table shows, there was a strong correlation between the third and fourth columns: the less energy a phone had at the mid frequencies, the more likely the reviewers were to describe it as accurate. Not listed in the table are the data for the bass and treble frequencies - I saw no clear-cut linear correlation in scatterplots regarding these.
There was some amount of uncertain facetry and some correlational mystery in the data, but won't go into that now. It might be apparent that there will be some degree of error in the dB numbers as they were based on estimations. Also, the results could well be due to random chance (small sample size) or a confounding variable that I failed to find.
I did a post-hoc by replacing the fourth column's values with the sum of just 2+3 kHz and found a reasonable correlation with the third column still. Doing the same with 3+4 kHz removed any appreciable correlation.
Edited by vid - 1/1/14 at 11:08am