The PerTh Neural Speech Watermarker is Resemble AI’s industry-leading audio watermarking technology for protecting voice content against unauthorized use and deepfake fraud. Built on the principles of psychoacoustics, PerTh embeds an inaudible watermark into audio files that persists through training on other speech synthesis models, allowing for reliable detection of IP ownership and copyright infringement.

How It Works

PerTh (named after Perceptual and Threshold) embeds watermarks into your audio content through a sophisticated process:

• Inaudible Embedding: The watermarker uses auditory masking principles to embed data in frequencies that are imperceptible to the human ear, ensuring audio quality remains unchanged.
• Persistence Through Training: Unlike traditional watermarks, PerTh remains persistent even when the audio is fed into another speech synthesis model for training, providing robust protection against unauthorized use.

About Resemble AI

Resemble AI is a pioneer in AI voice security and synthetic media detection. Our technology stack includes the PerTh Neural Watermarker, Resemble Detect (deepfake detection), and Identity services for comprehensive protection against AI fraud.