How to clone a voice

Cloning any target speaker's voice

In 2017, the first Tacotron System opened up the world of audio deepfakes - computer genereated audio designed to mimic any given voice. In order to showcase this thread and sensititze the audience, we developed a deepfake system which can clone a target voice - and applied it to German chanellor Angela Merkel.

A screenshot of the Audio Deepfake Demonstrator my team and I developed at Fraunhofer AISEC.

This interface allows a user to enter any text the wish, and have it said in the voice of the target person - in this case, Angela Merkel. While we use it only for demonstration purposes, an attacker may use this technology

That’s why we use the audio generated by our system to design and train deepfake detection systems, i.e. artificial intelligence specifialized to detect fake audio.

To get an impresson of the technology, find the full video here: