Case Notes 1

Animals equipped with the sense of hearing, for example humans, are commonly able to perceive sound sources spatially even though the sensory apparatus operates essentially in the time domain. This mapping of auditory space is managed by comparing inputs received at the two ears and analysing differences of level and timing.

Level differences are due in large part to the presence between the ears of the subject's head, which blocks and attenuates sound waves passing from one side to the other as long as their wavelength is of at most the same order of size as the head itself. For longer wavelengths, the interaural level differences (ILDs) due to the head's acoustic shadow are not significant, and this mechanism won't work.

Instead, for lower frequency sounds the brain uses synchronization differences due to the travel time from one ear to the other. The timescales involved are extremely short, and when this mechanism was originally proposed in the nineteenth century it was considered unlikely that the brain could distinguish timing differences with sufficient precision. However, a seminal experiment by Lord Rayleigh demonstrated that this does indeed occur and can be observed in the phenomenon of binaural beats.

The more familiar monaural beats are heard when two sounds of different frequencies are played simultaneously: the sound waves interfere with one another as they shift in and out of phase, causing a longer-period pulsation of the sound level. In the Rayleigh experiment, sounds of slightly different frequencies are presented separately to each ear through rubber tubes; even though the sound waves are not overlapping and therefore cannot physically cancel or reinforce one another, beats are still heard as a result of the brain noticing the phased time differences between the waves at each ear.

The processes behind this sensory feat remained mysterious until Lloyd Jeffress suggested an ingenious explanation in which arrays of coincidence detecting neurons in the brain are stimulated by the input from both ears but with incremental delays introduced along the pathways from the leading ear. A given neuron will fire only if it receives the signal pretty much simultaneously from both sides, which will only happen when the interaural time difference (ITD) is the same as the "known" delay on the input line to that neuron from the nearside ear. Each such neuron therefore effectively identifies a particular direction, and taken together they constitute an azimuthal map of the sound space.

This appealingly simple model has been widely accepted for over 50 years, and there is evidence that something resembling it does in fact operate in some species, notably the barn owl. However, recent results suggest it is not correct for mammals, nor even all birds (the barn owl's auditory adaptations are quite specialized). Although coincidence detection remains a key aspect of the process, neither the locally-coded map (ie, one neuron = one direction) nor the "delay lines" posited by Jeffress appear to exist.

Instead, the "delay" is really a kind of response suppression caused by pulsed release of an inhibitory neurotransmitter in parallel with the nerve impulses registering the sound -- exactly how this achieves the desired effect is not yet fully understood¹ -- while the mapping is probably non-local, which is to say the final direction is encoded in the patterns of firing of a number of different neurons, a much more efficient, but more complex, scheme.

An additional quirk, again reflecting a kind of neurological parsimony, is that -- unlike nearly everything else -- ITD recognition is not strictly contralateral (which is to say, managed by the hemisphere of the brain opposite the relevant ear). Instead, it actually switches from one side of the brain to the other as the ITDs change, even though both the sound source and the listener's perception of it remain on the same side.

The aspects of sound processing mentioned so far are relatively low-level, occurring long before signal information propagates up to the cerebral cortex. What happens at the higher levels, presumably leading to recognition and interpretation, is still pretty much unknown, but experiments with rats suggest that there is some plasticity in the responses of sound-associated neurons in the cortex. In other words, what the higher brain brings to the party is the ability to learn; not, perhaps, an earth-shattering revelation.

In one example, rats trained to recognise particular simple patterns of sound then exhibit some structure to their neural firing when given a complex, high-bandwidth sound stimulus, and that structure seems to have some statistical correlation with the learned sounds. It is essentially impossible to make this very specific -- you can't single out a neuron and say "that's the one!" -- and the models used are simplistic. In particular, there is an assumption of linearity even though it is likely that most of what is interesting about the neural behaviour is non-linear; but you have to start somewhere.

---

1 If this sounds handwaving, it is. If I were going to choose this case topic to focus on this term I'd want first and foremost to get a handle on what this glycinergic inhibition is about; but I'm not.
Posted by matt at October 30, 2006 10:13 PM