The rapid proliferation of AI-manipulated or generated audio deepfakes poses serious challenges to media integrity and election security. Current AI-driven detection solutions lack explainability and underperform in real-world settings. In this paper, we introduce novel explainability methods for state-of-the-art transformer-based audio deepfake detectors and open-source a novel benchmark for real-world generalizability. By narrowing the explainability gap between transformer-based audio deepfake detectors and traditional methods, our results not only build trust with human experts, but also pave the way for unlocking the potential of citizen intelligence to overcome the scalability issue in audio deepfake detection.
Real-world DeepFake videos often undergo various compression operations, resulting in a range of video qualities. These varying qualities diversify the pattern of forgery traces, significantly increasing the difficulty of DeepFake detection. To address this challenge, we introduce a new Dual Progressive Learning (DPL) framework for cross-quality DeepFake detection. We liken this task to progressively drilling for underground water, where low-quality videos require more effort than high-quality ones. To achieve this, we develop two sequential-based branches to "drill waters" with different efforts. The first branch progressively excavates the forgery traces according to the levels of video quality, i.e., time steps, determined by a dedicated CLIP-based indicator. In this branch, a Feature Selection Module is designed to adaptively assign appropriate features to the corresponding time steps. Considering that different techniques may introduce varying forgery traces within the same video quality, we design a second branch targeting forgery identifiability as complementary. This branch operates similarly and shares the feature selection module with the first branch. Our design takes advantage of the sequential model where computational units share weights across different time steps and can memorize previous progress, elegantly achieving progressive learning while maintaining reasonable memory costs. Extensive experiments demonstrate the superiority of our method for cross-quality DeepFake detection.
Due to the development of facial manipulation techniques in recent years deepfake detection in video stream became an important problem for face biometrics, brand monitoring or online video conferencing solutions. In case of a biometric authentication, if you replace a real datastream with a deepfake, you can bypass a liveness detection system. Using a deepfake in a video conference, you can penetrate into a private meeting. Deepfakes of victims or public figures can also be used by fraudsters for blackmailing, extorsion and financial fraud. Therefore, the task of detecting deepfakes is relevant to ensuring privacy and security. In existing approaches to a deepfake detection their performance deteriorates when multiple faces are present in a video simultaneously or when there are other objects erroneously classified as faces. In our research we propose to use geometric-fakeness features (GFF) that characterize a dynamic degree of a face presence in a video and its per-frame deepfake scores. To analyze temporal inconsistencies in GFFs between the frames we train a complex deep learning model that outputs a final deepfake prediction. We employ our approach to analyze videos with multiple faces that are simultaneously present in a video. Such videos often occur in practice e.g., in an online video conference. In this case, real faces appearing in a frame together with a deepfake face will significantly affect a deepfake detection and our approach allows to counter this problem. Through extensive experiments we demonstrate that our approach outperforms current state-of-the-art methods on popular benchmark datasets such as FaceForensics++, DFDC, Celeb-DF and WildDeepFake. The proposed approach remains accurate when trained to detect multiple different deepfake generation techniques.
Audio deepfake detection is crucial to combat the malicious use of AI-synthesized speech. Among many efforts undertaken by the community, the ASVspoof challenge has become one of the benchmarks to evaluate the generalizability and robustness of detection models. In this paper, we present Reality Defender's submission to the ASVspoof5 challenge, highlighting a novel pretraining strategy which significantly improves generalizability while maintaining low computational cost during training. Our system SLIM learns the style-linguistics dependency embeddings from various types of bonafide speech using self-supervised contrastive learning. The learned embeddings help to discriminate spoof from bonafide speech by focusing on the relationship between the style and linguistics aspects. We evaluated our system on ASVspoof5, ASV2019, and In-the-wild. Our submission achieved minDCF of 0.1499 and EER of 5.5% on ASVspoof5 Track 1, and EER of 7.4% and 10.8% on ASV2019 and In-the-wild respectively.
The rapid increase in deepfake technology has raised significant concerns about digital media integrity. Detecting deepfakes is crucial for safeguarding digital media. However, most standard image classifiers fail to distinguish between fake and real faces. Our analysis reveals that this failure is due to the model's inability to explicitly focus on the artefacts typically in deepfakes. We propose an enhanced architecture based on the GenConViT model, which incorporates weighted loss and update augmentation techniques and includes masked eye pretraining. This proposed model improves the F1 score by 1.71% and the accuracy by 4.34% on the Celeb-DF v2 dataset. The source code for our model is available at https://github.com/Monu-Khicher-1/multi-stage-learning
Detecting deepfakes has become an important task. Most existing detection methods provide only real/fake predictions without offering human-comprehensible explanations. Recent studies leveraging MLLMs for deepfake detection have shown improvements in explainability. However, the performance of pre-trained MLLMs (e.g., LLaVA) remains limited due to a lack of understanding of their capabilities for this task and strategies to enhance them. In this work, we empirically assess the strengths and weaknesses of MLLMs specifically in deepfake detection via forgery features analysis. Building on these assessments, we propose a novel framework called ${X}^2$-DFD, consisting of three core modules. The first module, Model Feature Assessment (MFA), measures the detection capabilities of forgery features intrinsic to MLLMs, and gives a descending ranking of these features. The second module, Strong Feature Strengthening (SFS), enhances the detection and explanation capabilities by fine-tuning the MLLM on a dataset constructed based on the top-ranked features. The third module, Weak Feature Supplementing (WFS), improves the fine-tuned MLLM's capabilities on lower-ranked features by integrating external dedicated deepfake detectors. To verify the effectiveness of this framework, we further present a practical implementation, where an automated forgery features generation, evaluation, and ranking procedure is designed for MFA module; an automated generation procedure of the fine-tuning dataset containing real and fake images with explanations based on top-ranked features is developed for SFS model; an external conventional deepfake detector focusing on blending artifact, which corresponds to a low detection capability in the pre-trained MLLM, is integrated for WFS module. Experiments show that our approach enhances both detection and explanation performance.
Advancements in artificial intelligence and machine learning have significantly improved synthetic speech generation. This paper explores diffusion models, a novel method for creating realistic synthetic speech. We create a diffusion dataset using available tools and pretrained models. Additionally, this study assesses the quality of diffusion-generated deepfakes versus non-diffusion ones and their potential threat to current deepfake detection systems. Findings indicate that the detection of diffusion-based deepfakes is generally comparable to non-diffusion deepfakes, with some variability based on detector architecture. Re-vocoding with diffusion vocoders shows minimal impact, and the overall speech quality is comparable to non-diffusion methods.
The rapid progress of Deepfake technology has made face swapping highly realistic, raising concerns about the malicious use of fabricated facial content. Existing methods often struggle to generalize to unseen domains due to the diverse nature of facial manipulations. In this paper, we revisit the generation process and identify a universal principle: Deepfake images inherently contain information from both source and target identities, while genuine faces maintain a consistent identity. Building upon this insight, we introduce DiffusionFake, a novel plug-and-play framework that reverses the generative process of face forgeries to enhance the generalization of detection models. DiffusionFake achieves this by injecting the features extracted by the detection model into a frozen pre-trained Stable Diffusion model, compelling it to reconstruct the corresponding target and source images. This guided reconstruction process constrains the detection network to capture the source and target related features to facilitate the reconstruction, thereby learning rich and disentangled representations that are more resilient to unseen forgeries. Extensive experiments demonstrate that DiffusionFake significantly improves cross-domain generalization of various detector architectures without introducing additional parameters during inference. Our Codes are available in https://github.com/skJack/DiffusionFake.git.
Image manipulation is rapidly evolving, allowing the creation of credible content that can be used to bend reality. Although the results of deepfake detectors are promising, deepfakes can be made even more complicated to detect through adversarial attacks. They aim to further manipulate the image to camouflage deepfakes' artifacts or to insert signals making the image appear pristine. In this paper, we further explore the potential of super-resolution attacks based on different super-resolution techniques and with different scales that can impact the performance of deepfake detectors with more or less intensity. We also evaluated the impact of the attack on more diverse datasets discovering that the super-resolution process is effective in hiding the artifacts introduced by deepfake generation models but fails in hiding the traces contained in fully synthetic images. Finally, we propose some changes to the detectors' training process to improve their robustness to this kind of attack.
Recent advances in Text-to-Speech (TTS) and Voice-Conversion (VC) using generative Artificial Intelligence (AI) technology have made it possible to generate high-quality and realistic human-like audio. This poses growing challenges in distinguishing AI-synthesized speech from the genuine human voice and could raise concerns about misuse for impersonation, fraud, spreading misinformation, and scams. However, existing detection methods for AI-synthesized audio have not kept pace and often fail to generalize across diverse datasets. In this paper, we introduce SONAR, a synthetic AI-Audio Detection Framework and Benchmark, aiming to provide a comprehensive evaluation for distinguishing cutting-edge AI-synthesized auditory content. SONAR includes a novel evaluation dataset sourced from 9 diverse audio synthesis platforms, including leading TTS providers and state-of-the-art TTS models. It is the first framework to uniformly benchmark AI-audio detection across both traditional and foundation model-based detection systems. Through extensive experiments, (1) we reveal the limitations of existing detection methods and demonstrate that foundation models exhibit stronger generalization capabilities, likely due to their model size and the scale and quality of pretraining data. (2) Speech foundation models demonstrate robust cross-lingual generalization capabilities, maintaining strong performance across diverse languages despite being fine-tuned solely on English speech data. This finding also suggests that the primary challenges in audio deepfake detection are more closely tied to the realism and quality of synthetic audio rather than language-specific characteristics. (3) We explore the effectiveness and efficiency of few-shot fine-tuning in improving generalization, highlighting its potential for tailored applications, such as personalized detection systems for specific entities or individuals.
The rapid advancement of deepfake technology poses a significant threat to digital media integrity. Deepfakes, synthetic media created using AI, can convincingly alter videos and audio to misrepresent reality. This creates risks of misinformation, fraud, and severe implications for personal privacy and security. Our research addresses the critical issue of deepfakes through an innovative multimodal approach, targeting both visual and auditory elements. This comprehensive strategy recognizes that human perception integrates multiple sensory inputs, particularly visual and auditory information, to form a complete understanding of media content. For visual analysis, a model that employs advanced feature extraction techniques was developed, extracting nine distinct facial characteristics and then applying various machine learning and deep learning models. For auditory analysis, our model leverages mel-spectrogram analysis for feature extraction and then applies various machine learning and deep learningmodels. To achieve a combined analysis, real and deepfake audio in the original dataset were swapped for testing purposes and ensured balanced samples. Using our proposed models for video and audio classification i.e. Artificial Neural Network and VGG19, the overall sample is classified as deepfake if either component is identified as such. Our multimodal framework combines visual and auditory analyses, yielding an accuracy of 94%.
With the significant advances in deep generative models for image and video synthesis, Deepfakes and manipulated media have raised severe societal concerns. Conventional machine learning classifiers for deepfake detection often fail to cope with evolving deepfake generation technology and are susceptible to adversarial attacks. Alternatively, invisible image watermarking is being researched as a proactive defense technique that allows media authentication by verifying an invisible secret message embedded in the image pixels. A handful of invisible image watermarking techniques introduced for media authentication have proven vulnerable to basic image processing operations and watermark removal attacks. In response, we have proposed a semi-fragile image watermarking technique that embeds an invisible secret message into real images for media authentication. Our proposed watermarking framework is designed to be fragile to facial manipulations or tampering while being robust to benign image-processing operations and watermark removal attacks. This is facilitated through a unique architecture of our proposed technique consisting of critic and adversarial networks that enforce high image quality and resiliency to watermark removal efforts, respectively, along with the backbone encoder-decoder and the discriminator networks. Thorough experimental investigations on SOTA facial Deepfake datasets demonstrate that our proposed model can embed a $64$-bit secret as an imperceptible image watermark that can be recovered with a high-bit recovery accuracy when benign image processing operations are applied while being non-recoverable when unseen Deepfake manipulations are applied. In addition, our proposed watermarking technique demonstrates high resilience to several white-box and black-box watermark removal attacks. Thus, obtaining state-of-the-art performance.
The adaptation of foundation models has significantly advanced environmental audio deepfake detection (EADD), a rapidly growing area of research. These models are typically fine-tuned or utilized in their frozen states for downstream tasks. However, the dimensionality of their representations can substantially lead to a high parameter count of downstream models, leading to higher computational demands. So, a general way is to compress these representations by leveraging state-of-the-art (SOTA) unsupervised dimensionality reduction techniques (PCA, SVD, KPCA, GRP) for efficient EADD. However, with the application of such techniques, we observe a drop in performance. So in this paper, we show that representation vectors contain redundant information, and randomly selecting 40-50% of representation values and building downstream models on it preserves or sometimes even improves performance. We show that such random selection preserves more performance than the SOTA dimensionality reduction techniques while reducing model parameters and inference time by almost over half.
In this study, we investigate multimodal foundation models (MFMs) for emotion recognition from non-verbal sounds. We hypothesize that MFMs, with their joint pre-training across multiple modalities, will be more effective in non-verbal sounds emotion recognition (NVER) by better interpreting and differentiating subtle emotional cues that may be ambiguous in audio-only foundation models (AFMs). To validate our hypothesis, we extract representations from state-of-the-art (SOTA) MFMs and AFMs and evaluated them on benchmark NVER datasets. We also investigate the potential of combining selected foundation model representations to enhance NVER further inspired by research in speech recognition and audio deepfake detection. To achieve this, we propose a framework called MATA (Intra-Modality Alignment through Transport Attention). Through MATA coupled with the combination of MFMs: LanguageBind and ImageBind, we report the topmost performance with accuracies of 76.47%, 77.40%, 75.12% and F1-scores of 70.35%, 76.19%, 74.63% for ASVP-ESD, JNV, and VIVAE datasets against individual FMs and baseline fusion techniques and report SOTA on the benchmark datasets.
Explainability in artificial intelligence is crucial for restoring trust, particularly in areas like face forgery detection, where viewers often struggle to distinguish between real and fabricated content. Vision and Large Language Models (VLLM) bridge computer vision and natural language, offering numerous applications driven by strong common-sense reasoning. Despite their success in various tasks, the potential of vision and language remains underexplored in face forgery detection, where they hold promise for enhancing explainability by leveraging the intrinsic reasoning capabilities of language to analyse fine-grained manipulation areas. As such, there is a need for a methodology that converts face forgery detection to a Visual Question Answering (VQA) task to systematically and fairly evaluate these capabilities. Previous efforts for unified benchmarks in deepfake detection have focused on the simpler binary task, overlooking evaluation protocols for fine-grained detection and text-generative models. We propose a multi-staged approach that diverges from the traditional binary decision paradigm to address this gap. In the first stage, we assess the models' performance on the binary task and their sensitivity to given instructions using several prompts. In the second stage, we delve deeper into fine-grained detection by identifying areas of manipulation in a multiple-choice VQA setting. In the third stage, we convert the fine-grained detection to an open-ended question and compare several matching strategies for the multi-label classification task. Finally, we qualitatively evaluate the fine-grained responses of the VLLMs included in the benchmark. We apply our benchmark to several popular models, providing a detailed comparison of binary, multiple-choice, and open-ended VQA evaluation across seven datasets. \url{https://nickyfot.github.io/hitchhickersguide.github.io/}
In this work, we explore Self-supervised Learning (SSL) as an auxiliary task to blend the texture-based local descriptors into feature modelling for efficient face analysis. Combining a primary task and a self-supervised auxiliary task is beneficial for robust representation. Therefore, we used the SSL task of mask auto-encoder (MAE) as an auxiliary task to reconstruct texture features such as local patterns along with the primary task for robust and unbiased face analysis. We experimented with our hypothesis on three major paradigms of face analysis: face attribute and face-based emotion analysis, and deepfake detection. Our experiment results exhibit that better feature representation can be gleaned from our proposed model for fair and bias-less face analysis.
The evolution of digital image manipulation, particularly with the advancement of deep generative models, significantly challenges existing deepfake detection methods, especially when the origin of the deepfake is obscure. To tackle the increasing complexity of these forgeries, we propose \textbf{Wavelet-CLIP}, a deepfake detection framework that integrates wavelet transforms with features derived from the ViT-L/14 architecture, pre-trained in the CLIP fashion. Wavelet-CLIP utilizes Wavelet Transforms to deeply analyze both spatial and frequency features from images, thus enhancing the model's capability to detect sophisticated deepfakes. To verify the effectiveness of our approach, we conducted extensive evaluations against existing state-of-the-art methods for cross-dataset generalization and detection of unseen images generated by standard diffusion models. Our method showcases outstanding performance, achieving an average AUC of 0.749 for cross-data generalization and 0.893 for robustness against unseen deepfakes, outperforming all compared methods. The code can be reproduced from the repo: \url{https://github.com/lalithbharadwajbaru/Wavelet-CLIP}
This paper introduces SpoofCeleb, a dataset designed for Speech Deepfake Detection (SDD) and Spoofing-robust Automatic Speaker Verification (SASV), utilizing source data from real-world conditions and spoofing attacks generated by Text-To-Speech (TTS) systems also trained on the same real-world data. Robust recognition systems require speech data recorded in varied acoustic environments with different levels of noise to be trained. However, current datasets typically include clean, high-quality recordings (bona fide data) due to the requirements for TTS training; studio-quality or well-recorded read speech is typically necessary to train TTS models. Current SDD datasets also have limited usefulness for training SASV models due to insufficient speaker diversity. SpoofCeleb leverages a fully automated pipeline we developed that processes the VoxCeleb1 dataset, transforming it into a suitable form for TTS training. We subsequently train 23 contemporary TTS systems. SpoofCeleb comprises over 2.5 million utterances from 1,251 unique speakers, collected under natural, real-world conditions. The dataset includes carefully partitioned training, validation, and evaluation sets with well-controlled experimental protocols. We present the baseline results for both SDD and SASV tasks. All data, protocols, and baselines are publicly available at https://jungjee.github.io/spoofceleb.
The perception module in autonomous vehicles (AVs) relies heavily on deep learning-based models to detect and identify various objects in their surrounding environment. An AV traffic sign classification system is integral to this module, which helps AVs recognize roadway traffic signs. However, adversarial attacks, in which an attacker modifies or alters the image captured for traffic sign recognition, could lead an AV to misrecognize the traffic signs and cause hazardous consequences. Deepfake presents itself as a promising technology to be used for such adversarial attacks, in which a deepfake traffic sign would replace a real-world traffic sign image before the image is fed to the AV traffic sign classification system. In this study, the authors present how a generative adversarial network-based deepfake attack can be crafted to fool the AV traffic sign classification systems. The authors developed a deepfake traffic sign image detection strategy leveraging hybrid quantum-classical neural networks (NNs). This hybrid approach utilizes amplitude encoding to represent the features of an input traffic sign image using quantum states, which substantially reduces the memory requirement compared to its classical counterparts. The authors evaluated this hybrid deepfake detection approach along with several baseline classical convolutional NNs on real-world and deepfake traffic sign images. The results indicate that the hybrid quantum-classical NNs for deepfake detection could achieve similar or higher performance than the baseline classical convolutional NNs in most cases while requiring less than one-third of the memory required by the shallowest classical convolutional NN considered in this study.
Face Forgery Detection (FFD), or Deepfake detection, aims to determine whether a digital face is real or fake. Due to different face synthesis algorithms with diverse forgery patterns, FFD models often overfit specific patterns in training datasets, resulting in poor generalization to other unseen forgeries. This severe challenge requires FFD models to possess strong capabilities in representing complex facial features and extracting subtle forgery cues. Although previous FFD models directly employ existing backbones to represent and extract facial forgery cues, the critical role of backbones is often overlooked, particularly as their knowledge and capabilities are insufficient to address FFD challenges, inevitably limiting generalization. Therefore, it is essential to integrate the backbone pre-training configurations and seek practical solutions by revisiting the complete FFD workflow, from backbone pre-training and fine-tuning to inference of discriminant results. Specifically, we analyze the crucial contributions of backbones with different configurations in FFD task and propose leveraging the ViT network with self-supervised learning on real-face datasets to pre-train a backbone, equipping it with superior facial representation capabilities. We then build a competitive backbone fine-tuning framework that strengthens the backbone's ability to extract diverse forgery cues within a competitive learning mechanism. Moreover, we devise a threshold optimization mechanism that utilizes prediction confidence to improve the inference reliability. Comprehensive experiments demonstrate that our FFD model with the elaborate backbone achieves excellent performance in FFD and extra face-related tasks, i.e., presentation attack detection. Code and models are available at https://github.com/zhenglab/FFDBackbone.