Probably. You can add in 'speaker' as a bit of metadata to the samples (this is what is meant by 'conditioning on') and teach it to speak like different people,

Probably. You can add in 'speaker' as a bit of metadata to the samples (this is what is meant by 'conditioning on') and teach it to speak like different people, so if you have a diverse sample of speakers and you add in 'accent' as another variable, it might well learn to disentangle individual speakers from their accents and then you can control generated accents by changing the metadata.