DeepL has introduced the Aure model to convey emotions in voice translation

The company DeepL has announced a new language model, Aure, designed to improve the quality of real-time voice translation. The technology is integrated into the DeepL Voice service and aims to maintain natural speech when switching languages.

A key feature of Aure is the system's ability to transfer not only the lexical meaning of words to the target language but also paralinguistic characteristics: intonation, emotional tone, accents, and the speaking speed of the original speaker. This allows for the conveyance of sarcasm, mood shifts, and other nuances of communication that were previously lost in automatic synthesis.

The development is intended to make the process of simultaneous translation less robotic. Unlike standard systems, which often use a monotonous voice, the new model analyzes the acoustic parameters of the original and adapts them to the target language model. This ensures a higher level of engagement and accuracy in conveying context during live dialogue.

For DeepL Voice users, the update means a shift toward higher-quality communication in professional and personal settings, where emotional subtext plays a decisive role. The tool is already available for testing as part of the company's updated platform, providing the ability to conduct meetings and negotiations while preserving the individual communication style of the participants.