In high-stakes settings, Machine Learning models that can provide predictions that are interpretable for humans are crucial. This is even more true with the advent of complex deep learning based models with a huge number of tunable parameters. Recently, prototype-based methods have emerged as a promising approach to make deep learning interpretable. We particularly focus on the analysis of deepfake videos in a forensics context. Although prototype-based methods have been introduced for the detection of deepfake videos, their use in real-world scenarios still presents major challenges, in that prototypes tend to be overly similar and interpretability varies between prototypes. This paper proposes a Visual Analytics process model for prototype learning, and, based on this, presents ProtoExplorer, a Visual Analytics system for the exploration and refinement of prototype-based deepfake detection models. ProtoExplorer offers tools for visualizing and temporally filtering prototype-based predictions when working with video data. It disentangles the complexity of working with spatio-temporal prototypes, facilitating their visualization. It further enables the refinement of models by interactively deleting and replacing prototypes with the aim to achieve more interpretable and less biased predictions while preserving detection accuracy. The system was designed with forensic experts and evaluated in a number of rounds based on both open-ended think aloud evaluation and interviews. These sessions have confirmed the strength of our prototype based exploration of deepfake videos while they provided the feedback needed to continuously improve the system.
The challenge in sourcing attribution for forgery faces has gained widespread attention due to the rapid development of generative techniques. While many recent works have taken essential steps on GAN-generated faces, more threatening attacks related to identity swapping or expression transferring are still overlooked. And the forgery traces hidden in unknown attacks from the open-world unlabeled faces still remain under-explored. To push the related frontier research, we introduce a new benchmark called Open-World DeepFake Attribution (OW-DFA), which aims to evaluate attribution performance against various types of fake faces under open-world scenarios. Meanwhile, we propose a novel framework named Contrastive Pseudo Learning (CPL) for the OW-DFA task through 1) introducing a Global-Local Voting module to guide the feature alignment of forged faces with different manipulated regions, 2) designing a Confidence-based Soft Pseudo-label strategy to mitigate the pseudo-noise caused by similar methods in unlabeled set. In addition, we extend the CPL framework with a multi-stage paradigm that leverages pre-train technique and iterative learning to further enhance traceability performance. Extensive experiments verify the superiority of our proposed method on the OW-DFA and also demonstrate the interpretability of deepfake attribution task and its impact on improving the security of deepfake detection area.
The malicious use and widespread dissemination of deepfake pose a significant crisis of trust. Current deepfake detection models can generally recognize forgery images by training on a large dataset. However, the accuracy of detection models degrades significantly on images generated by new deepfake methods due to the difference in data distribution. To tackle this issue, we present a novel incremental learning framework that improves the generalization of deepfake detection models by continual learning from a small number of new samples. To cope with different data distributions, we propose to learn a domain-invariant representation based on supervised contrastive learning, preventing overfit to the insufficient new data. To mitigate catastrophic forgetting, we regularize our model in both feature-level and label-level based on a multi-perspective knowledge distillation approach. Finally, we propose to select both central and hard representative samples to update the replay set, which is beneficial for both domain-invariant representation learning and rehearsal-based knowledge preserving. We conduct extensive experiments on four benchmark datasets, obtaining the new state-of-the-art average forgetting rate of 7.01 and average accuracy of 85.49 on FF++, DFDC-P, DFD, and CDF2. Our code is released at https://github.com/DeepFakeIL/DFIL.
Existing deepfake speech detection systems lack generalizability to unseen attacks (i.e., samples generated by generative algorithms not seen during training). Recent studies have explored the use of universal speech representations to tackle this issue and have obtained inspiring results. These works, however, have focused on innovating downstream classifiers while leaving the representation itself untouched. In this study, we argue that characterizing the long-term temporal dynamics of these representations is crucial for generalizability and propose a new method to assess representation dynamics. Indeed, we show that different generative models generate similar representation dynamics patterns with our proposed method. Experiments on the ASVspoof 2019 and 2021 datasets validate the benefits of the proposed method to detect deepfakes from methods unseen during training, significantly improving on several benchmark methods.
Audio deepfake detection (ADD) is the task of detecting spoofing attacks generated by text-to-speech or voice conversion systems. Spoofing evidence, which helps to distinguish between spoofed and bona-fide utterances, might exist either locally or globally in the input features. To capture these, the Conformer, which consists of Transformers and CNN, possesses a suitable structure. However, since the Conformer was designed for sequence-to-sequence tasks, its direct application to ADD tasks may be sub-optimal. To tackle this limitation, we propose HM-Conformer by adopting two components: (1) Hierarchical pooling method progressively reducing the sequence length to eliminate duplicated information (2) Multi-level classification token aggregation method utilizing classification tokens to gather information from different blocks. Owing to these components, HM-Conformer can efficiently detect spoofing evidence by processing various sequence lengths and aggregating them. In experimental results on the ASVspoof 2021 Deepfake dataset, HM-Conformer achieved a 15.71% EER, showing competitive performance compared to recent systems.
The rise of singing voice synthesis presents critical challenges to artists and industry stakeholders over unauthorized voice usage. Unlike synthesized speech, synthesized singing voices are typically released in songs containing strong background music that may hide synthesis artifacts. Additionally, singing voices present different acoustic and linguistic characteristics from speech utterances. These unique properties make singing voice deepfake detection a relevant but significantly different problem from synthetic speech detection. In this work, we propose the singing voice deepfake detection task. We first present SingFake, the first curated in-the-wild dataset consisting of 28.93 hours of bonafide and 29.40 hours of deepfake song clips in five languages from 40 singers. We provide a train/validation/test split where the test sets include various scenarios. We then use SingFake to evaluate four state-of-the-art speech countermeasure systems trained on speech utterances. We find these systems lag significantly behind their performance on speech test data. When trained on SingFake, either using separated vocal tracks or song mixtures, these systems show substantial improvement. However, our evaluations also identify challenges associated with unseen singers, communication codecs, languages, and musical contexts, calling for dedicated research into singing voice deepfake detection. The SingFake dataset and related resources are available at https://www.singfake.org/.
With the rise in manipulated media, deepfake detection has become an imperative task for preserving the authenticity of digital content. In this paper, we present a novel multi-modal audio-video framework designed to concurrently process audio and video inputs for deepfake detection tasks. Our model capitalizes on lip synchronization with input audio through a cross-attention mechanism while extracting visual cues via a fine-tuned VGG-16 network. Subsequently, a transformer encoder network is employed to perform facial self-attention. We conduct multiple ablation studies highlighting different strengths of our approach. Our multi-modal methodology outperforms state-of-the-art multi-modal deepfake detection techniques in terms of F-1 and per-video AUC scores.
Deepfake has recently raised a plethora of societal concerns over its possible security threats and dissemination of fake information. Much research on deepfake detection has been undertaken. However, detecting low quality as well as simultaneously detecting different qualities of deepfakes still remains a grave challenge. Most SOTA approaches are limited by using a single specific model for detecting certain deepfake video quality type. When constructing multiple models with prior information about video quality, this kind of strategy incurs significant computational cost, as well as model and training data overhead. Further, it cannot be scalable and practical to deploy in real-world settings. In this work, we propose a universal intra-model collaborative learning framework to enable the effective and simultaneous detection of different quality of deepfakes. That is, our approach is the quality-agnostic deepfake detection method, dubbed QAD . In particular, by observing the upper bound of general error expectation, we maximize the dependency between intermediate representations of images from different quality levels via Hilbert-Schmidt Independence Criterion. In addition, an Adversarial Weight Perturbation module is carefully devised to enable the model to be more robust against image corruption while boosting the overall model's performance. Extensive experiments over seven popular deepfake datasets demonstrate the superiority of our QAD model over prior SOTA benchmarks.
Generalisation -- the ability of a model to perform well on unseen data -- is crucial for building reliable deepfake detectors. However, recent studies have shown that the current audio deepfake models fall short of this desideratum. In this work we investigate the potential of pretrained self-supervised representations in building general and calibrated audio deepfake detection models. We show that large frozen representations coupled with a simple logistic regression classifier are extremely effective in achieving strong generalisation capabilities: compared to the RawNet2 model, this approach reduces the equal error rate from 30.9% to 8.8% on a benchmark of eight deepfake datasets, while learning less than 2k parameters. Moreover, the proposed method produces considerably more reliable predictions compared to previous approaches making it more suitable for realistic use.
Malicious use of deepfakes leads to serious public concerns and reduces people's trust in digital media. Although effective deepfake detectors have been proposed, they are substantially vulnerable to adversarial attacks. To evaluate the detector's robustness, recent studies have explored various attacks. However, all existing attacks are limited to 2D image perturbations, which are hard to translate into real-world facial changes. In this paper, we propose adversarial head turn (AdvHeat), the first attempt at 3D adversarial face views against deepfake detectors, based on face view synthesis from a single-view fake image. Extensive experiments validate the vulnerability of various detectors to AdvHeat in realistic, black-box scenarios. For example, AdvHeat based on a simple random search yields a high attack success rate of 96.8% with 360 searching steps. When additional query access is allowed, we can further reduce the step budget to 50. Additional analyses demonstrate that AdvHeat is better than conventional attacks on both the cross-detector transferability and robustness to defenses. The adversarial images generated by AdvHeat are also shown to have natural looks. Our code, including that for generating a multi-view dataset consisting of 360 synthetic views for each of 1000 IDs from FaceForensics++, is available at https://github.com/twowwj/AdvHeaT.
Due to the widespread use of smartphones with high-quality digital cameras and easy access to a wide range of software apps for recording, editing, and sharing videos and images, as well as the deep learning AI platforms, a new phenomenon of 'faking' videos has emerged. Deepfake algorithms can create fake images and videos that are virtually indistinguishable from authentic ones. Therefore, technologies that can detect and assess the integrity of digital visual media are crucial. Deepfakes, also known as deep learning-based fake videos, have become a major concern in recent years due to their ability to manipulate and alter images and videos in a way that is virtually indistinguishable from the original. These deepfake videos can be used for malicious purposes such as spreading misinformation, impersonating individuals, and creating fake news. Deepfake detection technologies use various approaches such as facial recognition, motion analysis, and audio-visual synchronization to identify and flag fake videos. However, the rapid advancement of deepfake technologies has made it increasingly difficult to detect these videos with high accuracy. In this paper, we aim to provide a comprehensive review of the current state of deepfake creation and detection technologies. We examine the various deep learning-based approaches used for creating deepfakes, as well as the techniques used for detecting them. Additionally, we analyze the limitations and challenges of current deepfake detection methods and discuss future research directions in this field. Overall, the paper highlights the importance of continued research and development in deepfake detection technologies in order to combat the negative impact of deepfakes on society and ensure the integrity of digital visual media.
Singing voice synthesis and singing voice conversion have significantly advanced, revolutionizing musical experiences. However, the rise of "Deepfake Songs" generated by these technologies raises concerns about authenticity. Unlike Audio DeepFake Detection (ADD), the field of song deepfake detection lacks specialized datasets or methods for song authenticity verification. In this paper, we initially construct a Chinese Fake Song Detection (FSD) dataset to investigate the field of song deepfake detection. The fake songs in the FSD dataset are generated by five state-of-the-art singing voice synthesis and singing voice conversion methods. Our initial experiments on FSD revealed the ineffectiveness of existing speech-trained ADD models for the task of song deepFake detection. Thus, we employ the FSD dataset for the training of ADD models. We subsequently evaluate these models under two scenarios: one with the original songs and another with separated vocal tracks. Experiment results show that song-trained ADD models exhibit a 38.58% reduction in average equal error rate compared to speech-trained ADD models on the FSD test set.
The rise of deepfake images, especially of well-known personalities, poses a serious threat to the dissemination of authentic information. To tackle this, we present a thorough investigation into how deepfakes are produced and how they can be identified. The cornerstone of our research is a rich collection of artificial celebrity faces, titled DeepFakeFace (DFF). We crafted the DFF dataset using advanced diffusion models and have shared it with the community through online platforms. This data serves as a robust foundation to train and test algorithms designed to spot deepfakes. We carried out a thorough review of the DFF dataset and suggest two evaluation methods to gauge the strength and adaptability of deepfake recognition tools. The first method tests whether an algorithm trained on one type of fake images can recognize those produced by other methods. The second evaluates the algorithm's performance with imperfect images, like those that are blurry, of low quality, or compressed. Given varied results across deepfake methods and image changes, our findings stress the need for better deepfake detectors. Our DFF dataset and tests aim to boost the development of more effective tools against deepfakes.
Face forgery instances involving celebrities are on the rise, owing to the ease with which their large quantity of videos may be accessible on the Internet, world leaders particularly. While current face manipulation detectors have achieved impressive results on several open datasets, which incorporate persons with various identities, they show performance degradation on these high-quality ones targeting at celebrities. What is more, these online videos usually undergo compression processing, marking the detection task harder. Besides, more face manipulation techniques arise for celebrities other than face-swap, such as lip-synchronize and image-animation, with which most works have not been concerned. This paper proposes a dual stream learning facial and speaking patterns method to protect celebrities against deepfakes. We design an action unit module based on facial action coding system along with an Action Unit Transformer (AUT) to exploit facial expressions embeddings. Besides, our method's dual stream architecture utilizes a Temporal Convolutional Network (TCN) to extract lip motion pattern and learns the relatedness between facial and speaking patterns. Our method could protect the person of interest (POI) against deepfakes in an end-to-end manner. Extensive experiments show that our method achieves better performance and has a higher resistance to video compression than state-of-the-art detection models.
There are growing implications surrounding generative AI in the speech domain that enable voice cloning and real-time voice conversion from one individual to another. This technology poses a significant ethical threat and could lead to breaches of privacy and misrepresentation, thus there is an urgent need for real-time detection of AI-generated speech for DeepFake Voice Conversion. To address the above emerging issues, the DEEP-VOICE dataset is generated in this study, comprised of real human speech from eight well-known figures and their speech converted to one another using Retrieval-based Voice Conversion. Presenting as a binary classification problem of whether the speech is real or AI-generated, statistical analysis of temporal audio features through t-testing reveals that there are significantly different distributions. Hyperparameter optimisation is implemented for machine learning models to identify the source of speech. Following the training of 208 individual machine learning models over 10-fold cross validation, it is found that the Extreme Gradient Boosting model can achieve an average classification accuracy of 99.3% and can classify speech in real-time, at around 0.004 milliseconds given one second of speech. All data generated for this study is released publicly for future research on AI speech detection.
Current anti-spoofing and audio deepfake detection systems use either magnitude spectrogram-based features (such as CQT or Melspectrograms) or raw audio processed through convolution or sinc-layers. Both methods have drawbacks: magnitude spectrograms discard phase information, which affects audio naturalness, and raw-feature-based models cannot use traditional explainable AI methods. This paper proposes a new approach that combines the benefits of both methods by using complex-valued neural networks to process the complex-valued, CQT frequency-domain representation of the input audio. This method retains phase information and allows for explainable AI methods. Results show that this approach outperforms previous methods on the "In-the-Wild" anti-spoofing dataset and enables interpretation of the results through explainable AI. Ablation studies confirm that the model has learned to use phase information to detect voice spoofing.
This paper introduces our system designed for Track 2, which focuses on locating manipulated regions, in the second Audio Deepfake Detection Challenge (ADD 2023). Our approach involves the utilization of multiple detection systems to identify splicing regions and determine their authenticity. Specifically, we train and integrate two frame-level systems: one for boundary detection and the other for deepfake detection. Additionally, we employ a third VAE model trained exclusively on genuine data to determine the authenticity of a given audio clip. Through the fusion of these three systems, our top-performing solution for the ADD challenge achieves an impressive 82.23% sentence accuracy and an F1 score of 60.66%. This results in a final ADD score of 0.6713, securing the first rank in Track 2 of ADD 2023.
The exploitation of Deepfake techniques for malicious intentions has driven significant research interest in Deepfake detection. Deepfake manipulations frequently introduce random tampered traces, leading to unpredictable outcomes in different facial regions. However, existing detection methods heavily rely on specific forgery indicators, and as the forgery mode improves, these traces become increasingly randomized, resulting in a decline in the detection performance of methods reliant on specific forgery traces. To address the limitation, we propose Recap, a novel Deepfake detection model that exposes unspecific facial part inconsistencies by recovering faces and enlarges the differences between real and fake by mapping recovered faces. In the recovering stage, the model focuses on randomly masking regions of interest (ROIs) and reconstructing real faces without unpredictable tampered traces, resulting in a relatively good recovery effect for real faces while a poor recovery effect for fake faces. In the mapping stage, the output of the recovery phase serves as supervision to guide the facial mapping process. This mapping process strategically emphasizes the mapping of fake faces with poor recovery, leading to a further deterioration in their representation, while enhancing and refining the mapping of real faces with good representation. As a result, this approach significantly amplifies the discrepancies between real and fake videos. Our extensive experiments on standard benchmarks demonstrate that Recap is effective in multiple scenarios.
A deepfake is a photo or video of a person whose image has been digitally altered or partially replaced with an image of someone else. Deepfakes have the potential to cause a variety of problems and are often used maliciously. A common usage is altering videos of prominent political figures and celebrities. These deepfakes can portray them making offensive, problematic, and/or untrue statements. Current deepfakes can be very realistic, and when used in this way, can spread panic and even influence elections and political opinions. There are many deepfake detection strategies currently in use but finding the most comprehensive and universal method is critical. So, in this survey we will address the problems of malicious deepfake creation and the lack of universal deepfake detection methods. Our objective is to survey and analyze a variety of current methods and advances in the field of deepfake detection.
Deepfakes are becoming increasingly credible, posing a significant threat given their potential to facilitate fraud or bypass access control systems. This has motivated the development of deepfake detection methods, in which deep learning models are trained to distinguish between real and synthesized footage. Unfortunately, existing detectors struggle to generalize to deepfakes from datasets they were not trained on, but little work has been done to examine why or how this limitation can be addressed. Especially, those single-modality deepfake images reveal little available forgery evidence, posing greater challenges than detecting deepfake videos. In this work, we present the first empirical study on the generalizability of deepfake detectors, an essential goal for detectors to stay one step ahead of attackers. Our study utilizes six deepfake datasets, five deepfake image detection methods, and two model augmentation approaches, confirming that detectors do not generalize in zero-shot settings. Additionally, we find that detectors are learning unwanted properties specific to synthesis methods and struggling to extract discriminative features, limiting their ability to generalize. Finally, we find that there are neurons universally contributing to detection across seen and unseen datasets, suggesting a possible path towards zero-shot generalizability.