CoVoST 2 is released, a large-scale multilingual speech translation corpus covering translations from 21 languages into English and from English into 15 languages, which represents the largest open dataset available to date from total volume and language coverage perspective.