Update the dataset to achieve more alignment between sentence and homograph

Fix a dataset issue

Add the homograph dataset

More data clean-up (fixed plurals yielding an extra word in LibriSpeech data)

Initial commit