DeepL Releases Aure Real-Time Speech Translation Model
Understand the story
On November 11, 2026, The Decoder reported that DeepL released its new Aure voice model. The model carries the speaker's tone, emotion, emphasis, and speaking pace into the target language, changing the previous approach that relied on a static voice sample captured at the start of a conversation, which could not track emotional shifts or sarcasm. Aure continuously analyzes the translation and the original audio during conversation; during training, timbre, speech melody, and content are separated to enable independent transfer, and accents are removed through reinforcement learning. Aure runs automatically across DeepL Voice's online meetings, face-to-face conversations, and the DeepL API. No further follow-up reports have been seen.
AI overview of reporting · Updated 36 min前
Reporting timeline
Explore different aspects of the story through its reporting.
- The DecoderDeepL wants real-time translation to finally capture sarcasm and mood shifts
DeepL launched its new Aure model, which carries a speaker's tone, emotion, emphasis and speaking pace into the target language, replacing the previous approach that relied on static voice samples captured at the start of a session and could not detect emotional shifts or sarcasm. Aure continuously analyzes translations and original audio during conversations, was trained by separating timbre, speech melody and content for independent transfer, and uses reinforcement learning to remove accents. It runs automatically in DeepL Voice's online meetings, face-to-face conversations and DeepL API.
Story heat trend
Not enough continuous observations to show a trend yet.