Real-Time Voice Translation Is Wonderful and a Voice-Actor Problem
ElevenLabs and Mistral both shipped real-time voice translation this week, across dozens of languages. As someone who records things, I think it is genuinely useful. As someone who respects consent, I think it is a landmine.
A voice is identity rendered as a file. When you can move that identity across a language the speaker does not speak, the line between 'helpful' and 'impersonation' gets very thin, very fast.
The tools will be used for good things: a clinic reaching a patient, a small business selling across a border. They will also be used to fake someone's consent in a language they never agreed to.
My rule is unchanged and non-negotiable: label synthetic speech as synthetic, and never clone a voice without written permission. The tech is wonderful. The ethics are not automatic, and the person shipping the pipeline is the only safeguard that exists.
