Please can you clarify what is an acceptable use of metadata for this competition. The host has already implied that it could be used to remove problematic utterances from the data (e.g. #characters is unusual for the duration).
Can we use the metdata provided for training or inference? Are there any fields that are off limits?