How to deal with AI-generated audio contributions?

Common Voice
1

Today, while listening to some clips, I found 2 or 3 that sounded like they were generated by an AI model: they were very clear and the rhythm and pronunciation were very consistent.
I can’t be sure they were AI-generated, but there are probably some online tools we could use to check. Is there a way to download the clips I reported?
I reported them using the “Other” category, but I don’t know the exact policy (though I assume the project only accepts human-generated audio).
I don’t have specific suggestions; I’m opening this thread so we can discuss the subject. This is probably not the first time someone has done something that harms the project.

One idea would be to train a model to classify audio as human- or AI-generated and use it as a filter before releasing the dataset.

2

Here is another topic on the same subject. If you search for TTS you will find more.

3

I think it would be worth checking these clips before removing them, since some synthetic voices can sound surprisingly natural. Maybe we could flag suspicious recordings for review and add a clear note about AI-generated audio rather than relying entirely on an automated detector.