chunked/nltk_chunking/model_doc_mms/chunk_5.txt

Experiments show that our multilingual speech recognition model more than halves the word error rate of 
Whisper on 54 languages of the FLEURS benchmark while being trained on a small fraction of the labeled data.