chunked/nltk_chunking/model_doc_audio-spectrogram-transformer/chunk_15.txt

You can check ast/src/get_norm_stats.py to see how
the authors compute the stats for a downstream dataset.