Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/facebookresearch/audioseal/llms.txt

Use this file to discover all available pages before exploring further.

​
Introduction to AudioSeal

AudioSeal is a state-of-the-art audio watermarking method designed for proactive detection of AI-generated audio content. Developed by Meta’s FAIR team, it provides an efficient solution for embedding imperceptible watermarks into audio and detecting them with unprecedented speed.

​
What is AudioSeal?

AudioSeal introduces a novel approach to audio watermarking using localized watermarking at the sample level (1/16,000 of a second). The system jointly trains two neural network components:
  • Generator: Embeds an imperceptible watermark into audio signals
  • Detector: Identifies watermark fragments in long or edited audio files
AudioSeal works well with multiple sampling rates including 16 kHz, 24 kHz, 44.1 kHz, and 48 kHz, making it versatile for various audio applications.

​
Key Features

AudioSeal stands out from existing watermarking solutions with several key advantages:

​
Localized Watermarking

  • Sample-level precision at 1/16,000 of a second
  • Enables detection in edited or spliced audio content
  • Works seamlessly across different sampling rates

​
Minimal Audio Quality Impact

  • Imperceptible watermarks that maintain audio fidelity
  • Uses novel perceptual loss during training
  • Designed to preserve the listening experience

​
Robust Against Modifications

  • Resistant to audio compression and re-encoding
  • Survives noise addition and various audio edits
  • Maintains detection accuracy through transformations

​
Ultra-Fast Detection

  • Two orders of magnitude faster than existing models
  • Single-pass detection algorithm
  • Ideal for large-scale and real-time applications

​
Optional Secret Messages

  • Support for 16-bit secret messages (65,536 possible values)
  • Can identify model versions or add metadata
  • Detection works independently of message content

​
Use Cases

AudioSeal is designed for various audio watermarking scenarios:

​
AI-Generated Audio Detection

Identify and track AI-generated speech and audio content proactively, helping to prevent misuse of synthetic voices.

​
Content Attribution

Embed watermarks to trace the source and authenticity of audio content, enabling proper attribution and copyright protection.

​
Media Verification

Verify whether audio has been generated by AI systems, supporting fact-checking and content moderation efforts.

​
Real-Time Applications

Process streaming audio with the fast detection system, enabling live monitoring and verification at scale.

​
Model Version Tracking

Use the optional 16-bit message to identify which model version generated specific audio content.

​
Technical Highlights

Starting from AudioSeal 0.2+, audio is not resampled internally. Users are responsible for providing audio at the appropriate sample rate for their use case.
AudioSeal v0.2.0 brings several improvements:
  • Streaming support for real-time watermarking (requires Python ≥3.10)
  • Enhanced compatibility with PyTorch 2.6+
  • Full MIT license including model weights for commercial use
  • Training code available for custom models

​
Research and Resources

AudioSeal is based on peer-reviewed research presented at ICML 2024: Paper: Proactive Detection of Voice Cloning with Localized Watermarking
Authors: Robin San Roman, Pierre Fernandez, Hady Elsahar, Alexandre Défossez, Teddy Furon, Tuan Tran

​
Additional Resources

The team also develops other open-source watermarking solutions:
  • WMAR - Autoregressive watermarking models for images
  • Video Seal - Open and efficient video watermarking
  • WAM - Watermark any images with localization

​
Next Steps

1

Install AudioSeal

Get started by installing the AudioSeal package via pip or from source. View installation instructions →
2

Try the Quickstart

Follow our quickstart guide to watermark your first audio file in minutes. View quickstart guide →
3

Explore the API

Learn about the full API capabilities for advanced use cases. View API reference →

Build docs developers (and LLMs) love

Get started for freeTalk to us