Two questions about transcription challenge

Ego4D Challenges
1

I want to ask two questions regarding speech transcription challenge.

(1) Is it okay to use OpenAI’s whisper model (Introducing Whisper) for our challenge? I know that it’s a publicly available model, but want to make sure.

(2) If so, do you normalise our submission json file before evaluation? Or do I have to submit the json file after normalising with english.glm provided in the github repo?

Thank you!

2

  1. I am not sure how you use Whisper as part of your solution, but please be sure you comply with Challenge Official Rules https://ego4d-interactive-fig1.s3.eu-west-2.amazonaws.com/tc.pdf especially the part “Intellectual Property”.

  2. Before scoring the submissions, our evaluation script will itself lowercase the words in the transcript and apply the English GLM.