Feature request: Voice swap

[lochstar]

If possible, it'd be handy to have an existing recording that can then be swapped with a model.