(back to SpatialLocalizationPaper)

Most of the previous phenomena could be seen as relatively straightforward applications of geometry and physics. The observance of binaural masking level differences (MLD) comes somewhat surprisingly by comparison, but turns out to be enormously useful. A MLD difference arises when the ears give the brain two different chances to examine the relationship between signal and noise. This is best demonstrated by a pair of experiments: When a tone is overlaid (dioticly) on white noise just softly enough to be masked, inverting the phase of the tone in one ear lowers the masking threshold enough for it to become audible again; when the tone is fed to one ear with sufficient masking noise, adding the same noise to the other ear again makes the tone audible. Thus, we conclude that the brain is comparing interaural phase information. Sounds that are identical at both ears can be filtered out, or more generally, sounds with one phase relationship can be filtered with respect to sounds with another. Such effects can be observed for music, speech, and even transients, primarily in the low frequencies where the MLD can reach 15 dB (Moore, 168-9).

The above experiments suggest an immediate, if not exactly quotidian application: in precise analogy with the second example, pilots can improve the audibility of sounds in their earphones over the background noise by reversing one speaker's polarity (Gulick, 341). On a more commonplace scale, though, we can use MLD to explain the so-called "cocktail party effect," wherein we discern different conversations merely by focusing our attention despite large amount of noise (including the conversations on which we were previously eavesdropping). The effect is clearly a binaural one, as plugging one ear will quickly reveal. The ability to merely localize each source using previously described methods won't work -- we'll see that the HaasEffect actually makes apparent localization a poor way to discriminate between complex sounds. It turns out that the masking noise must be in the same critical band as the target sound, so an ability to filter speech with other speech is not surprising. In fact, there is a measurable intelligibility level difference correlating our ability to understand speech at certain levels of proficiency with different amounts of masking. Unsurprisingly, the limiting case of 0% ILD is identical to the MLD of the same intensities, and thus we equate the two effects (Gelfand, 313-5).

A few models have attempted to explain MLDs. The fact that differences >2dB require frequencies lower than 1500 Hz immediately suggests a dependence on the interaural phase detection described in BinauralLocalization. However, a complete theory is still somewhat lacking. See Moore, 172 for the findings of Webster-Jeffress and Durlach.

onward to HaasEffect