Fast and Effective Copy-Move Detection of Digital Audio Based on Auto Segment

Huang, Xinchao; Liu, Zihan; Lu, Wei; Li, Hongmei; Xiang, Shijun

doi:10.4018/978-1-7998-3025-2.ch011

Cited by 14 publications

(9 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Finally, the cross-entropy loss is computed between the similarity region mask and the label (a binary image of size 1 × 128 × 256, where forged region pixels are 255 and normal region pixels are 0, normalized to [0, 1] during the calculation). The loss function is defined as Equation (6).…”

Section: Methodsmentioning

confidence: 99%

“…In Equation (6), n represents the total number of features, which is 1 × 128 × 256, where P(x i ) represents the value of the label and Q(x i ) represents the predicted value of the localization model.…”

Section: Methodsmentioning

confidence: 99%

“…Most of these methods compute the feature similarity between all audio segments. However, Liu et al [5,6] adopt a different approach by rapidly sorting the features of all segments. They then only calculate the similarity between one voiced segment and its adjacent voiced segments, reducing the time complexity of computing similarity between audio segments from O(n2) to O(nlog2n), with no significant degradation in detection performance.…”

Section: Introductionmentioning

confidence: 99%

“…The existing passive evidence-gathering methods typically involve three steps: The first step is to segment the audio into word or phrase segments, remove silent segments, and then extract various features from each segment. For audio segmentation, more researchers [2][3][4][5][6][7][8][9][10] use Voice Activity Detection (VAD) to divide the audio into silent and audio segments and [11,12] use Normalized Low-Frequency Energy Ratio (NLFER) to divide the audio into silent and audio segments and [13] to divide the audio into equal time periods. Then, various features are extracted from the segmented audio segments, such as singular feature vector [2], Mel scale frequency cepstral coefficient (MFCC) [3,7], fundamental sequence [4], discrete Fourier transform (DFT) coefficient [5][6][7], gamma filter [7], histogram [8,9], color autocorrelation map [10], tone similarity [11], tone sequence [12], format sequence [12], etc.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

An Audio Copy-Move Forgery Localization Model by CNN-Based Spectral Analysis

Zhao,

Zhang,

Wang

et al. 2024

Applied Sciences

View full text Add to dashboard Cite

In audio copy-move forgery forensics, existing traditional methods typically first segment audio into voiced and silent segments, then compute the similarity between voiced segments to detect and locate forged segments. However, audio collected in noisy environments is difficult to segment and manually set, and heuristic similarity thresholds lack robustness. Existing deep learning methods extract features from audio and then use neural networks for binary classification, lacking the ability to locate forged segments. Therefore, for locating audio copy-move forgery segments, we have improved deep learning methods and proposed a robust localization model by CNN-based spectral analysis. In the localization model, the Feature Extraction Module extracts deep features from Mel-spectrograms, while the Correlation Detection Module automatically decides on the correlation between these deep features. Finally, the Mask Decoding Module visually locates the forged segments. Experimental results show that compared to existing methods, the localization model improves the detection accuracy of audio copy-move forgery by 3.0–6.8%and improves the average detection accuracy of forged audio with post-processing attacks such as noise, filtering, resampling, and MP3 compression by over 7.0%.

show abstract

Section: Methodsmentioning

confidence: 99%

Section: Methodsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

An Audio Copy-Move Forgery Localization Model by CNN-Based Spectral Analysis

Zhao,

Zhang,

Wang

et al. 2024

Applied Sciences

View full text Add to dashboard Cite

show abstract

“…Audio forensics has been intensively studied [19][20][21][22]. Audio reverberation significantly affects the quality of an audio recording and small differences can indicate alterations in the original recordings.…”

Section: Introductionmentioning

confidence: 99%

Blackman–Tukey spectral estimation and electric network frequency matching from power mains and speech recordings

Karantaidis

Kotropoulos

2021

IET Signal Processing

View full text Add to dashboard Cite

Forensic applications exploit electric network frequency (ENF) as a fingerprint to determine multimedia content authenticity, as well as the time and region of multimedia recording. ENF is present at a nominal frequency of 50/60 Hz and its harmonics. Strong interference due to speech content deteriorates ENF estimation accuracy. Herein, the authors propose a non-parametric approach for ENF estimation, which incorporates a customised lag window design into the Blackman-Tukey spectral estimation method. Leakage reduction is formulated as a problem of energy maximisation within the main lobe of the spectral window. The proposed approach is compared to state-of-the-art methods for ENF estimation. Maximum correlation coefficient and minimum standard deviation of errors are employed to measure ENF estimation accuracy. Hypothesis testing is performed to determine whether the improvements in ENF estimation accuracy of the proposed approach over the state-of-the-art methods are statistically significant. Experimental results and statistical tests indicate that the proposed approach improves ENF estimation against many state-of-the-art methods.This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.

show abstract