Cloning a Voice Without Retraining the Model

Voice cloning, as Amazon's machine-learning blog describes it, doesn't require training a model on a new speaker: the system generates new speech in a target speaker's voice from a short reference recording. The workflow is two steps — supply the reference recording, then supply the new text, and the model speaks that text in the reference speaker's voice, still without retraining. The vendor's case for running the model yourself is control: a self-hosted, publicly available model lets you manage cost and keep audio data inside your own AWS environment. For anyone weighing that against a hosted service, the last point is the real product — a recording of someone's voice is the kind of data teams would rather not hand over.