Low-dimensional interference of mid-level sound statistics predicts human speech recognition in natural environmental noise

This is a supporting dataset for the manuscript "Low-dimensional interference of mid-level sound statistics predicts human speech recognition in natural environmental noise". The dataset itself is comprised of three psychoacoustic experiments that investigate human speech recognition in differing natural enviornments. In the first experiment, (n=18) participants recognize spoken digit triplets in the presence of 11 natural backgrounds, and acoustically perturbed variants that whiten the the modulation content (Phase Randomized, PR) or the spectrum content (Spectrum Equalized, SE) of the sound. In the second experiment, (n=16) participants recognize spoken digit triplets in the presence of the Jackhammer Sound or the 8 Speaker Babble sound, that have been perturbed by gradually added texture statistics (McDermott 2011). In the third experiment, (n=9) participants recognize spoken digit triplets in the presence of 11 natural backgrounds at 7 different, signal-to-noise ratios. The supported data will be able to replicate the psychoacoustic results presented in the paper, in addition to serving as the input for the logistic regression model used in subsequent analysis. The repository contains Audio Files (.wav format) and Behavioral Data (MATLAB .mat format).