ElevenLabs is embedding Google DeepMind's SynthID (an invisible watermark) into the audio it produces - undetectable to the ear, survives cutting and compression. Plus a free detector anyone can use to check where a sound came from.
- ElevenLabs is embedding Google DeepMind's SynthID into generated audio: a watermark undetectable to the ear (June 25).
- Survives cutting, speed changes, compression, format conversion and metadata removal.
- A free Audio Detector checks whether a sound was made by ElevenLabs; starting with free Text-to-Speech.
Run a clip with their voice through the detector, and a name comes out. ElevenLabs announced a partnership with Google DeepMind - an invisible watermark, SynthID, embedded directly into the audio they generate. Plus a free Audio Detector anyone can use to check whether a given sound is theirs.
There's real stakes here, and it hits me personally - we use ElevenLabs exactly for our assistant's voice. It's done right: if the machine is talking, it should show it's a machine. Honesty built into the sound itself.
Provenance - what's real and what's generated - stops being a matter of good faith and becomes a technical function. Good that a leading provider went and made it a standard, not an extra.