Skip to main content
Every clip generated on a supported model automatically carries an inaudible watermark that identifies it as AI-generated by Cartesia. It helps you and your users verify where audio came from, and supports your compliance needs as AI transparency requirements evolve.

Check whether audio is watermarked

Public detector. Anyone can upload a clip at cartesia.ai/watermark-detector to check for a Cartesia watermark. Supports FLAC, M4A, MP3, MP4, MPEG, MPGA, OGA, OGG, WAV and WebM files between 3 seconds and 5 minutes long, up to 32 MB. Detection API (paid plans). Check audio programmatically, for example to screen uploads or verify clips.
Reading the result. Detection is statistical. detected: true means the clip very likely contains a Cartesia watermark. False positives are rare but possible. detected: false means no Cartesia watermark was found. The audio could still be AI-generated by another provider, or heavily edited.

FAQs

Yes. It’s applied automatically to all audio generated by Sonic-3.6 or newer models.
No. Watermarking is always on and can’t be disabled at any level.
Yes. Watermarking covers every generation endpoint: WebSocket, SSE, and Bytes.
Yes. It’s robust to common audio transformations, including compression, resampling, and conversion.
Anyone can use the public detector. Customers on paid plans can also use the Detection API.
Not yet. Today, the watermark identifies audio as Cartesia-generated. Per-customer attribution for paid plans is planned for a future release.
No. There’s no perceptible impact on either.