It's infuriating seeing non-profits gatekeep datasets that were compiled with grant money. At least Elsevier doesn't present itself as a charity.

I was recently trying to get my hands on the Switchboard and Fisher conversational speech datasets. Both were funded by DARPA grants, and maintained by the non-profit LDC, which charges you thousands of dollars for access (and no discounts for individual researchers) - that is, if they'll even pay attention to you without a .edu email address. And both are standard corpora in the field of audio NLP, which makes replicating studies impossible.

Sadly, I couldn't find any way to pirate the datasets - they're too niche. So I applaud the authors for sticking it to Worldcat and scraping their data.