Effect of auditory-nerve response variability on estimates of tuning curves.

Near-Poisson variability in auditory-nerve (AN) responses limits the accuracy of automated tuning-curve algorithms. Here, a typical adaptive tuning-curve algorithm was used with a physiologically realistic AN model with and without the inclusion of neural randomness. Response randomness produced variability in Q(10) estimates that was nearly as large as in AN data. Results suggest that it is sufficient for AN models to specify frequency selectivity based on mean Q(10) values at each characteristic frequency (CF). Errors in estimates of CF, which decreased from +/-0.2 octaves at low frequencies to +/-0.05 octaves at high frequencies, are significant for studies of spatiotemporal coding.