For our paper "Warning words in a warming world : Central bank communication and climate change", my co-authors and I created the most comprehensive dataset of central bank speeches ever collected. This dataset is now open access and will be updated regularly 

[update for 2024-2025 is currently ongoing!]

The CBS dataset contains 35,487 speeches, building on the BIS dataset (18,045 speeches) augmented with the systematic web-scraping of central banks' websites (15,435 new speeches) and additional work in central bank archives (2,007 original speeches). It includes over 5000 speeches in original non-English language, as well as their machine-generated translation to English. It also includes a broader set of metadata, including speakers' gender and position.