ElevenLabs, an artificial intelligence (AI) voice-focused startup, said on the 10th that it will offer its latest dubbing AI model, "Dubbing v2," via an application programming interface (API).
"Dubbing v2" is a model that reproduces the original voice's emotions as-is, and it was first released in May as a user interface (UI) version. With this API release, developers and corporations can now directly link and integrate it into their own products and workflows (work processes).
The model adopts an audio-to-audio architecture that converts voice directly to voice. Unlike the conventional "speech recognition → translation → speech synthesis (clone voice)" method, it generates voice output directly from the input voice, allowing it to faithfully reflect the original's emotional expression, the company said.
"Dubbing v2" supports 92 languages, including Korean and Japanese. ElevenLabs expects the model to be used for localizing game characters, producing multilingual marketing videos, and distributing films and drama overseas. An ElevenLabs official said, "When K-content expands overseas, we expect two-way use, including high-quality multilingual dubbing and Korean dubbing of leading overseas content."