The Hurricane Challenge Dataset

The Hurricane Challenge was an international evaluation of intelligibility-enhancing speech modifications that took place in 2013 at Interspeech. The dataset consists of stereo files containing clean speech (channel 1) and noise (channel 2). The purpose of the Challenge is to modify the clean speech only in order to make it more intelligibility in the presence of the corresponding noise sample, without changing root-mean-square level, and within certain constraints on changes in duration. Two different types of noise, stationary speech-shaped noise, and competing speech, are provided, each at three signal-to-noise ratios. The Challenge is described in this Interspeech article.  

The dataset consists of two zip files, one for each masker (cs.zip for competing speech, ssn.zip for speech-shaped noise).