Meta AudioSeal

Properties

category: watermarking
maturity: emerging
approach: Uses a novel perceptual loss function inspired by auditory masking to embed imperceptible watermarks for localized detection (sample-level precision).
access: open-source
owns: Watermarks AI-generated audio with imperceptible, localized signals so synthetic speech can be flagged at the sample level; an open-source Meta Research project.
distinctFrom: Unlike SynthID which uses a spectrogram-based approach for audio, AudioSeal operates with a single-pass detector optimized for localized, fast, sample-level detection.
source: Meta Research (GitHub)
notes: Meta researchers claim it is the first audio watermarking system designed specifically for localized detection and operates up to two orders of magnitude faster than previous models (unverified).

category: watermarking
maturity: emerging
approach: Uses a novel perceptual loss function inspired by auditory masking to embed imperceptible watermarks for localized detection (sample-level precision).
access: open-source
owns: Watermarks AI-generated audio with imperceptible, localized signals so synthetic speech can be flagged at the sample level; an open-source Meta Research project.
distinctFrom: Unlike SynthID which uses a spectrogram-based approach for audio, AudioSeal operates with a single-pass detector optimized for localized, fast, sample-level detection.
source: Meta Research (GitHub)
notes: Meta researchers claim it is the first audio watermarking system designed specifically for localized detection and operates up to two orders of magnitude faster than previous models (unverified).