On the last day of physiology, BDA injections were made along the recording paths to estimate recording sites. We measured average firing rates, Z scores, precision,

and selectivity from the responses of individual neurons. Z scores were measured as (driven firing rate − baseline firing rate)/(SD of baseline firing rate). We quantified trial-to-trial precision by first computing the shuffled autocorrelogram using the spiking responses to individual songs ( Joris et al., 2006). The shuffled autocorrelogram quantifies the propensity of neurons to fire spikes across multiple presentations of the same stimulus at varying lags. The correlation index is the shuffled autocorrelogram value at a lag of 0 ms, and it indicates SRT1720 clinical trial the propensity to fire spikes at the same time (±0.5 ms) each time the stimulus is presented. To quantify selectivity, we first determined the number of songs that drove at least one significant spiking event. Significant spiking events were defined by two criteria: (1) the smoothed PSTH (binned at 1 ms and smoothed with a 20 ms Hanning window) had to exceed baseline activity (p < 0.05), and (2) during this duration, spiking activity had to occur on >50% of trials. Selectivity was then quantified as 1 − (n/15), where n was the number of songs (out of 15) that drove at least one significant spiking event. To quantify population sparseness, we computed the fraction of neurons that produced significant

spiking events during every 63 ms epoch, using a sliding window. We then quantified the fraction of neurons active during each window, with low values indicating PAK6 higher levels of sparseness. To create population PSTHs, we first computed Microbiology inhibitor the PSTH of each individual

neuron within a population in response to a single song, smoothed with a 5 ms Hanning window. We then averaged the PSTHs of every neuron in a population, without normalizing. To quantify the degree to which neural responses to auditory scenes reflected the individual song within the scene, we computed an extraction index using the PSTHs to a scene at a particular SNR, as well as the PSTHs to the song and chorus components of that scene. From these PSTHs we computed two correlation coefficients: Rsong was the correlation between the song and scene PSTHs and Rchor was the correlation coefficient between the scene and the chorus PSTHs. The extraction index was defined as (Rsong − Rchor)/(Rsong + Rchor). Other methods for quantifying the extraction index from the PSTHs or from single spike trains produced qualitatively and quantitatively similar results. STRFs were calculated from the spiking responses to individual vocalization and the corresponding spectrograms using a generalized linear model, as previously described (Calabrese et al., 2011). We validated the predictive quality of each STRF by predicting the response to a song not used during estimation. We then calculated the correlation coefficient between the predicted and actual PSTHs.