Selectivity for interaural phase and/or time differences across carrier waveforms in modeled neurons resembles the impulses of auditory filters used to model cochlear filters. Furthermore, the ‘labels’ of the same neurons, under the edge model of spatial hearing, should span little more than ~0.2 cycles if phase ambiguity is to be avoided at the sub-maximal activity level (i.e., from -0.2 to 0 and 0 to 0.2 cycles, respectively, in the left and right hemispheres).
The panel below shows a single log-normal neuron in the left hemisphere (blue line) or range of ‘labeled’ neurons (±0.2 cycles, blue lines), displayed atop the impulse of an auditory filter (gray line) having a selectable center frequency (100 to 1,600 Hz).
Given the resemblance of auditory filter impulses and selectivity for IPD across carrier waveforms, selectivity for interaural phase and/or time differences across the envelopes of waveforms may be similarly shaped by the envelopes of the same impulses.
Single log-normal neuron in the left hemisphere (blue line), or range of ‘labeled’ neurons (blue lines), selective for interaural phase and/or time differences across the envelopes of waveforms, displayed atop the impulse of an auditory filter (gray line) having a selectable center frequency (100 to 12,800 Hz).
Selectivity for interaural differences across the envelopes of waveforms could likely be derived directly from the envelopes of modeled auditory filters. However, the log-normal neurons shown above are the same as those selective for carrier waveforms in the prior panel, scaled by the ratio of each neuron’s half-maximal width (e.g., 0.428 cycles) over each frequency-specific impulses’s half-maximal width (ranging, e.g., from 1.810 cycles at 100 Hz to 5.229 cycles at 12,800 Hz).
The main reason for using this scaled approach is to simplify modeling, allowing a single set of neurons to be used for IPDs corresponding to carrier waveforms as well as the envelopes of the same waveforms. To simplify modeling even further, IPDs derived from the envelopes of waveforms may, alternatively, be divided by the same frequency-specific ratios and neurons from the previous panel may be used without scaling.