@alice@w7voa But you need the voice recordings alongside their metadata to train and validate your speech and intent recognition models.
So you have a choice between:
Shitty experience for minors/people who sound like minors
Storing voice data of minors/people who sound like minors
#2 can be somewhat mitigated with an automated deletion policy, though you have to come up with a strategy to mitigate continual data loss of your golden training/test set.