I want to generate WSJ0-kspeakers-mixture. Where and how can I get the WSJ0 dataset?

Question

I'm trying to create the WSJ0 mixtures, but I'm having trouble. I downloaded the create-speaker-mixtures.zip and spatialize_wsj0-mix.zip from https://www.merl.com/demos/deep-clustering. However, the create_wav_2speakers.mat says "This script assumes that WSJ0's wv1 sphere files have already been converted to wav files, using the original folder structure under wsj0/, e.g., 11-1.1/wsj0/si_tr_s/01t/01to030v.wv1 is converted to wav and % stored in YOUR_PATH/wsj0/si_tr_s/01t/01to030v.wav" and such. I have no idea what these are supposed to be. They weren't part of the downloaded files. Simply running the code creates empty files names "mix_2_spk_min_tr_1", "mix_2_spk_min_tr_2" and "mix_2_spk_min_tr_mix", which are of no use to me. I've tried looking at the other folder, spatialize_wsj0-mix.zip, but that requires "the original wsj0-mix dataset, which can be built using the scripts available at create-speaker-mixtures.zip". I tried looking elsewhere for help, but all the codes require some "directory containing the WSJ0 dataset (containing the wsj0/ folder)". So... where do I get that original dataset? Do I have to pay for it somewhere? It seems everyone assumes I have that already, but I simply can't seem to find anywhere that tells me where I can download them.

Answer 1

Yes, you have to pay for it here: https://catalog.ldc.upenn.edu/LDC93S6A because it is a licensed dataset