Summary

This repository serves as a curated aggregation of research progress and resources for audio deepfake detection (ADD). It targets researchers and practitioners by consolidating relevant papers, datasets, and publicly available code, aiming to streamline the study of speech deepfake detection techniques.

How It Works

The project functions as a comprehensive knowledge base, systematically organizing information on audio deepfake detection. It compiles survey papers, lists prominent repositories, details advanced audio large language models, and categorizes a vast array of datasets, preprocessing techniques, enhancement methods, and feature extraction approaches. This structured aggregation provides a broad overview of the research landscape and facilitates comparative analysis of detection methodologies.

Quick Start & Requirements

This repository is a curated collection of research resources and does not offer direct installation or execution commands. Users are directed to external papers and code repositories for specific implementations.

Highlighted Details

• Extensive compilation of survey papers covering audio deepfake detection and multimodal AI-generated content detection.
• Detailed catalog of audio large language models (e.g., AudioLM, VALL-E, VoiceBox) with their respective capabilities and publication dates.
• Comprehensive listing of audio deepfake detection datasets, categorized by attack types (TTS, Replay, VC, etc.), including size, language, and year.
• Information on audio preprocessing, enhancement methods, and various feature extraction techniques (handcrafted, hybrid, end-to-end).
• Performance metrics (EER, t-DCF) for numerous detection methods are presented, highlighting state-of-the-art results.

Maintenance & Community

Suggestions and error reports are welcomed via email (xuyuxiong2022@email.szu.edu.cn). No other community channels or explicit maintenance schedules are detailed.

Licensing & Compatibility

No specific open-source license is declared. The project states its content is sourced from journals and the internet for communication and learning. Aggregated materials are subject to their original copyrights, with a policy to remove infringing content upon notification. This suggests a non-commercial, research-oriented use case and potential restrictions on redistribution or commercial application of sourced materials.

Limitations & Caveats

This is a resource aggregator, not a deployable software tool. Its utility is confined to research and learning, relying on external links for actual code and datasets. Users must independently verify the licensing and accessibility of all sourced materials, and the project's copyright policy necessitates careful consideration of the legal status of aggregated content.