Voice Identity Disentanglement

DeepL has unveiled Aure, a new model for its Voice real-time translation product that carries the original speaker's tone, emotion, emphasis and speaking pace into the target language instead of relying on a static voice sample captured at conversation start. It decouples voice identity, speech melody and content during training and uses reinforcement learning to suppress accents, rolling out automatically across Online Meetings, In-Person Conversations and the API.