AES Journal Forum

Listening to headphones gives me a total different sound-image as from well placed loudspeakers in an adequate room. With loudspeakers the sound seems to stem from the space between the speakers, the wall behind them, the flour and the ceiling. The ability of 'filling' this space with sound, is one of the quality aspects of loudspeakers.
With headphones the sound comes from behind as long as the headphones have been 'connected' to your head. Anyway, in my opinion headphones never will be able to reproduce this aspect of speakers, so to me judging loudspeakers via headphones at least leaves this point outside of consideration.

The loudspeakers interaction with the room should theoretically be captured perfectly within a properly acquired binaural room impulse response (including all wall & ceiling reflections, and also create proper phantom source perception). The Localizations errors you described will vanish, when using an interactive, i.e. head-tracked auralization of tested loudspeakers (see Sean Olive & the BRS system, and our own works for instance).

Real issues are the missing reproduction of the non-linear behaviour of the loudspeakers within an auralization (mentioned in Hiekkanens paper). Another factor is the individuality aspect, which was nicely tackled in the paper, though the proposed approach remains unfeasable with interactive simulations, as compensation measurements had to be done for all possible head movements.

We also experienced the reported colouration problems mainly due to improper HF compensation, but what is more, to our subjects LF-deviations of the same order were perceptible. To me it seems to be a problem to convincingly reproduce LF sound (below say 80Hz) with headphones within auralizations. The impact sound a large wideband speaker delivers to the listener is not reproducible via (the most often used open type) headphones.

Still, even it was feasible to produce individualized, interactive auralizations of loudspeakers the significant deviations in the HF & LF Range remain to be solved.