Skip to content

AudioSeal

Watermark AI speech for fast detection

Visit official site

What AudioSeal is

The technique is described as a localised watermark for AI generated speech, pairing a generator with a single-pass detector for long audio.

What you can do with it

  • Add a watermark to an audio signal
  • Detect watermarked segments after editing
  • Read a secret message from the detector

Pros & cons

✓ What we like

  • A 16-bit secret message can be encoded in the watermark
  • Training code is provided for custom watermark models

! What to watch out for

  • Detection works on watermarked audio segments
  • The detector output is a probability per sample

FAQ

What are the two parts?

A generator and a detector trained jointly.

Can I build my own model?

Training code is provided.

Last reviewed: 2026-11-15

More AI audio tools tools

View all →

How we review