chunked/nltk_chunking/model_doc_xls_r/chunk_5.txt

For speech recognition, XLS-R improves over the best known prior work on BABEL, MLS, CommonVoice as well as
VoxPopuli, lowering error rates by 14-34% relative on average.