Audio generated by ElevenLabs gains an invisible watermark

ElevenLabs integrates Google DeepMind SynthID invisible watermarking into its free Text-to-Speech audio to ensure traceability against transformations.

Recognizing an AI voice by ear is becoming increasingly unreliable. To enhance the traceability of its audio content, ElevenLabs is partnering with Google DeepMind to integrate SynthID into the content generated by its platform.

This digital watermark, inaudible to the user, is embedded directly into the audio file and must remain detectable even after common transformations such as cutting, compression, format conversion, or sound acceleration. The watermarking begins with Text-to-Speech content generated by free users, before a planned extension to all ElevenLabs audio content. In parallel, the company is launching a free Audio Detector, allowing users to verify if an audio clip originates from its tools.

ElevenLabs describes this layer as a complement to its existing traceability systems and to provenance standards like C2PA, in a context where several jurisdictions are already requiring machine-identifiable synthetic content.