In this case the determining factor is that the original submission was not the verbatim speech that the speaker gave, and the speaker himself complained that that some of the transcription was inaccurate.

This then caused significant meta-discussion about the inaccuracy of the transcription. For the other comments that were about the content itself, we can't be sure if those comments were made in response to parts of the talk that were accurately or inaccurately transcribed.

We understand there may be overlapping comments, but that's a price we have to pay given the other considerations. No decision would be perfect, and it's not the same as usual "dupe" scenarios. We have to be a bit flexible when the situation is different from the norm in important ways.