For discrimination to work, spike trains elicited from multiple presentations of the same song must be more similar to each other than spike trains elicited from different songs. We therefore computed the similarity of each two spike trains by various methods, all dependent on a temporal scale factor that gauges the precision of spike timing. Conventional cluster analysis allows one to group the spike trains and assign them to one of the stimuli. The correctness of this assignment can be evaluated by the mutual information, giving a measure of the discrimination performance.
The following results were obtained: (1) Conspecific songs can be discriminated after a few hundred milliseconds. (2) Discrimination works best if spikes are evaluated at time scales of 10 msec. (3) Although the length of repetitive subunits helps in discriminating the songs, sufficient information must be contained in the structure of these subunits. Thus spike trains of auditory receptors carry enough information to enable a representation of the songs at high resolution.