⚡️ Introducing DETECT-2B, the most advanced deepfake detection model


Protect your IP

Resemble’s Neural Watermarker can prevent your data from being used by AI models by tracing it back to the source.

Read more about PerTH, our Deep Learning Watermarker

Balenciaga Logo
Leo Burnett Logo
Netflix Logo

Over 200,135 AI voices generate more than 2,000,000 minutes of audio per month on Resemble!

Introducing PerTh: A Combination of Perceptual and Threshold

PerTh is a sophisticated deep neural network watermarker developed by Resemble AI that embeds imperceptible data into generated speech content, creating an “invisible watermark”. This watermark is difficult to remove and provides a secure method to verify if a specific audio clip was generated by Resemble, thereby providing an added layer of protection against the misuse of AI-generated audio.

frequency-amplitude space demonstrating "auditory masking"

AI Watermarker

Harnessing the Power of Psychoacoustics

Frequency Sensitivity Exploitation

Our hearing sensitivity varies across different frequencies. By understanding this fundamental concept, PerTh ingeniously embeds more data into those frequencies to which we’re less sensitive. The result? A watermark that’s securely tucked away in the audio, undetectable to the human ear, yet entirely retrievable when needed.

Auditory Masking Unleashed

PerTh cleverly uses the principle of “auditory masking” to its advantage. When a loud sound (or masker sound) is played, quieter sounds nearby become imperceptible. PerTh leverages this effect to place watermarked data into these “masked” areas, ensuring the main audio content remains undisturbed.

Balancing Data Security with Audio Quality

With PerTh Watermarker, we’ve found the perfect balance between robust data security and uncompromised audio quality. By exploiting the principles of psychoacoustics, we provide an innovative watermarking solution that ensures your audio content’s authenticity without impacting the listening experience.

For Developers

Technologist on Apple iOS 16.4 API Ready.

Integrating our cutting-edge PerTh Watermarker into your systems has never been easier, thanks to our user-friendly API. We’ve designed our API with simplicity and flexibility in mind, providing a seamless experience for businesses of all sizes and complexities.

Resemble AI Detect API

How does the PerTh Watermarker work?

PerTh, short for ‘Perceptual Threshold’, uses psychoacoustic principles to encode data within audio content. It identifies frequencies that are inaudible to human listeners and encodes data into these regions. This method ensures that the data payload is difficult to corrupt with common audio manipulations.

Can the watermark be removed or tampered with?

The watermark is tightly coupled with the audio content, making it difficult to remove without significantly degrading the audio quality. Furthermore, any attempts at tampering can be detected by analyzing the watermark.