The Earnings 21 dataset ( also referred to as earnings21 ) is a 39-hour corpus of earnings calls containing entity dense speech from nine different financial sectors. This corpus is intended to benchmark automatic speech recognition (ASR) systems in the wild with special attention towards named entity recognition (NER).
In this repo, we provided the transcoded files to 16KHz with the formatted text. This limits the utility of the original dataset due to the restrictions… See the full description on the dataset page:
https://huggingface.co/datasets/Revai/earnings21.
