Manipulated Regions Localization For Partially Deepfake Audio: A Survey

View on arXiv ← Back to list

Authors: Jiayi He, Jiangyan Yi, Jianhua Tao, Siding Zeng, Hao Gu

Published: 2025-06-17 10:51:34+00:00

AI Summary

This survey provides the first comprehensive overview of partially deepfake audio manipulated region localization tasks. It systematically introduces existing methods, datasets, evaluation metrics, and challenges, highlighting future research directions and potential trends in this field.

Abstract

With the development of audio deepfake techniques, attacks with partially deepfake audio are beginning to rise. Compared to fully deepfake, it is much harder to be identified by the detector due to the partially cryptic manipulation, resulting in higher security risks. Although some studies have been launched, there is no comprehensive review to systematically introduce the current situations and development trends for addressing this issue. Thus, in this survey, we are the first to outline a systematic introduction for partially deepfake audio manipulated region localization tasks, including the fundamentals, branches of existing methods, current limitations and potential trends, providing a revealing insight into this scope.

Key findings

Significant progress has been made in partially deepfake audio localization, with improvements in both accuracy and the ability to locate manipulated regions. However, challenges remain in addressing insufficient localization accuracy for complex scenarios and the need for stronger physical evidence supporting localization results. Future trends include focusing on inconsistencies between manipulated and non-manipulated regions and leveraging LLMs for evidence provision.

Approach

The survey categorizes existing approaches to partially deepfake audio localization into four types: frame-level authenticity, boundary perception, frame-level inconsistency, and multi-modality fusion. It analyzes the strengths and weaknesses of each approach, comparing their performance across various datasets.

Datasets

PartialSpoof, HAD, ADD2022Track2, ADD2023Track2, Psynd, LAV-DF, AV-Deepfake1M, LlamaPartialSpoof

Model(s)

LCNN, ResNet, SENet, LSTM, CQCC, MFCC, LFCC, Wav2vec, WavLM, BiLSTM, CRNN, SELCNN, SPF, TDL, W-TDL, CFPRF, BAM, AGO, PET, GNCL, UMMAFormer, BA-TFD, BA-TFD+

Author countries

China

← Previous