WER mismatch in EN ASR model

[ankitapasad]

I am using the fine-tuned ASR EN models provided in the repo. I have two questions:

1. The dict.en_char.txt file the downloaded model folder does not have an apostrophe symbol, whereas from what I understand in the paper apostrophes are retained. The numeral "1" though in the dict.en_char looks out of place, is that a typo?
2. When I evaluate the EN model on the EN dev set (no language model) I get a WER of 13.9, as opposed to the published result of 15.5. For this part I replaced the "1" in the dict.en_char to an apostrophe "`". (note: I preprocessed the data to remove all punctuations except apostrophe).

Any insights on these issues are appreciated!