Gaurav Bharaj

Sepehr Dehdashtian, Jacob H Seidman, Vishnu N Boddeti et al.

Audio deepfake detection (ADD) models are critical for countering the malicious use of text-to-speech (TTS) models. Evaluating and strengthening ADD models requires developing datasets that span the space of generated audio and highlight high-error regions. Existing dataset development strategies face two challenges: (i) manual collection, and (ii) inefficient discovery of blind spots in the ADD models. To address these challenges, we propose FoeGlass, the first black-box automated red-teaming method for ADDs, which effectively discovers ADD failure modes in the space of generated audio underexplored by state-of-the-art deepfake benchmarks. FoeGlass uses the in-context learning capabilities of an LLM to explore the input space of a TTS model, generating audio samples that fool the target ADD using only black-box access to all components. By using a carefully designed context based on diversity measurements, FoeGlass mitigates the common problem of mode collapse in automated red-teaming systems. Empirical evaluations on several open-source ADD and TTS models demonstrate that data generated from FoeGlass substantially improves the false negative rates over unconditional sampling baselines and recent spoofing datasets by up to 94%, while requiring no manual supervision. Furthermore, we show that the attacks generated by FoeGlass are transferable across different target ADDs, demonstrating its broad applicability and ease of use for the automated red teaming of ADD systems. Finally, fine-tuning ADD models on FoeGlass-generated samples notably enhances the robustness of the detectors (up 41%).

Yi Zhu, Surya Koppisetti, Trang Tran et al. · uw

Audio deepfake detection (ADD) is crucial to combat the misuse of speech synthesized from generative AI models. Existing ADD models suffer from generalization issues, with a large performance discrepancy between in-domain and out-of-domain data. Moreover, the black-box nature of existing models limits their use in real-world scenarios, where explanations are required for model decisions. To alleviate these issues, we introduce a new ADD model that explicitly uses the StyleLInguistics Mismatch (SLIM) in fake speech to separate them from real speech. SLIM first employs self-supervised pretraining on only real samples to learn the style-linguistics dependency in the real class. The learned features are then used in complement with standard pretrained acoustic features (e.g., Wav2vec) to learn a classifier on the real and fake classes. When the feature encoders are frozen, SLIM outperforms benchmark methods on out-of-domain datasets while achieving competitive results on in-domain data. The features learned by SLIM allow us to quantify the (mis)match between style and linguistic content in a sample, hence facilitating an explanation of the model decision.

Xingzhe He, Gaurav Bharaj, David Ferman et al.

Supervised keypoint localization methods rely on large manually labeled image datasets, where objects can deform, articulate, or occlude. However, creating such large keypoint labels is time-consuming and costly, and is often error-prone due to inconsistent labeling. Thus, we desire an approach that can learn keypoint localization with fewer yet consistently annotated images. To this end, we present a novel formulation that learns to localize semantically consistent keypoint definitions, even for occluded regions, for varying object categories. We use a few user-labeled 2D images as input examples, which are extended via self-supervision using a larger unlabeled dataset. Unlike unsupervised methods, the few-shot images act as semantic shape constraints for object localization. Furthermore, we introduce 3D geometry-aware constraints to uplift keypoints, achieving more accurate 2D localization. Our general-purpose formulation paves the way for semantically conditioned generative modeling and attains competitive or state-of-the-art accuracy on several datasets, including human faces, eyes, animals, cars, and never-before-seen mouth interior (teeth) localization tasks, not attempted by the previous few-shot methods. Project page: https://xingzhehe.github.io/FewShot3DKP/}{https://xingzhehe.github.io/FewShot3DKP/

Chuhan Chen, Matthew O'Toole, Gaurav Bharaj et al.

High-quality reconstruction of controllable 3D head avatars from 2D videos is highly desirable for virtual human applications in movies, games, and telepresence. Neural implicit fields provide a powerful representation to model 3D head avatars with personalized shape, expressions, and facial parts, e.g., hair and mouth interior, that go beyond the linear 3D morphable model (3DMM). However, existing methods do not model faces with fine-scale facial features, or local control of facial parts that extrapolate asymmetric expressions from monocular videos. Further, most condition only on 3DMM parameters with poor(er) locality, and resolve local features with a global neural field. We build on part-based implicit shape models that decompose a global deformation field into local ones. Our novel formulation models multiple implicit deformation fields with local semantic rig-like control via 3DMM-based parameters, and representative facial landmarks. Further, we propose a local control loss and attention mask mechanism that promote sparsity of each learned deformation field. Our formulation renders sharper locally controllable nonlinear deformations than previous implicit monocular approaches, especially mouth interior, asymmetric expressions, and facial details.

Chirag Goel, Surya Koppisetti, Ben Colman et al.

Vision transformers (ViT) have made substantial progress for classification tasks in computer vision. Recently, Gong et. al. '21, introduced attention-based modeling for several audio tasks. However, relatively unexplored is the use of a ViT for audio spoof detection task. We bridge this gap and introduce ViTs for this task. A vanilla baseline built on fine-tuning the SSAST (Gong et. al. '22) audio ViT model achieves sub-optimal equal error rates (EERs). To improve performance, we propose a novel attention-based contrastive learning framework (SSAST-CL) that uses cross-attention to aid the representation learning. Experiments show that our framework successfully disentangles the bonafide and spoof classes and helps learn better classifiers for the task. With appropriate data augmentations policy, a model trained on our framework achieves competitive performance on the ASVSpoof 2021 challenge. We provide comparisons and ablation studies to justify our claim.

Nisarg A. Shah, Gaurav Bharaj

We present a novel algorithm to reduce tensor compute required by a conditional image generation autoencoder without sacrificing quality of photo-realistic image generation. Our method is device agnostic, and can optimize an autoencoder for a given CPU-only, GPU compute device(s) in about normal time it takes to train an autoencoder on a generic workstation. We achieve this via a two-stage novel strategy where, first, we condense the channel weights, such that, as few as possible channels are used. Then, we prune the nearly zeroed out weight activations, and fine-tune the autoencoder. To maintain image quality, fine-tuning is done via student-teacher training, where we reuse the condensed autoencoder as the teacher. We show performance gains for various conditional image generation tasks: segmentation mask to face images, face images to cartoonization, and finally CycleGAN-based model over multiple compute devices. We perform various ablation studies to justify the claims and design choices, and achieve real-time versions of various autoencoders on CPU-only devices while maintaining image quality, thus enabling at-scale deployment of such autoencoders.

David Ferman, Gaurav Bharaj

We present a novel method for multi image domain and multi-landmark definition learning for small dataset facial localization. Training a small dataset alongside a large(r) dataset helps with robust learning for the former, and provides a universal mechanism for facial landmark localization for new and/or smaller standard datasets. To this end, we propose a Vision Transformer encoder with a novel decoder with a definition agnostic shared landmark semantic group structured prior, that is learnt, as we train on more than one dataset concurrently. Due to our novel definition agnostic group prior the datasets may vary in landmark definitions and domains. During the decoder stage we use cross- and self-attention, whose output is later fed into domain/definition specific heads that minimize a Laplacian-log-likelihood loss. We achieve state-of-the-art performance on standard landmark localization datasets such as COFW and WFLW, when trained with a bigger dataset. We also show state-of-the-art performance on several varied image domain small datasets for animals, caricatures, and facial portrait paintings. Further, we contribute a small dataset (150 images) of pareidolias to show efficacy of our method. Finally, we provide several analysis and ablation studies to justify our claims.

Yue Zhang, Ben Colman, Xiao Guo et al.

State-of-the-art deepfake detection approaches rely on image-based features extracted via neural networks. While these approaches trained in a supervised manner extract likely fake features, they may fall short in representing unnatural `non-physical' semantic facial attributes -- blurry hairlines, double eyebrows, rigid eye pupils, or unnatural skin shading. However, such facial attributes are easily perceived by humans and used to discern the authenticity of an image based on human common sense. Furthermore, image-based feature extraction methods that provide visual explanations via saliency maps can be hard to interpret for humans. To address these challenges, we frame deepfake detection as a Deepfake Detection VQA (DD-VQA) task and model human intuition by providing textual explanations that describe common sense reasons for labeling an image as real or fake. We introduce a new annotated dataset and propose a Vision and Language Transformer-based framework for the DD-VQA task. We also incorporate text and image-aware feature alignment formulation to enhance multi-modal representation learning. As a result, we improve upon existing deepfake detection models by integrating our learned vision representations, which reason over common sense knowledge from the DD-VQA task. We provide extensive empirical results demonstrating that our method enhances detection performance, generalization ability, and language-based interpretability in the deepfake detection task.

Petr Grinberg, Ankur Kumar, Surya Koppisetti et al.

Adding explanations to audio deepfake detection (ADD) models will boost their real-world application by providing insight on the decision making process. In this paper, we propose a relevancy-based explainable AI (XAI) method to analyze the predictions of transformer-based ADD models. We compare against standard Grad-CAM and SHAP-based methods, using quantitative faithfulness metrics as well as a partial spoof test, to comprehensively analyze the relative importance of different temporal regions in an audio. We consider large datasets, unlike previous works where only limited utterances are studied, and find that the XAI methods differ in their explanations. The proposed relevancy-based XAI method performs the best overall on a variety of metrics. Further investigation on the relative importance of speech/non-speech, phonetic content, and voice onsets/offsets suggest that the XAI results obtained from analyzing limited utterances don't necessarily hold when evaluated on large datasets.

Sepehr Dehdashtian, Mashrur M. Morshed, Jacob H. Seidman et al.

Synthetic image detectors (SIDs) are a key defense against the risks posed by the growing realism of images from text-to-image (T2I) models. Red teaming improves SID's effectiveness by identifying and exploiting their failure modes via misclassified synthetic images. However, existing red-teaming solutions (i) require white-box access to SIDs, which is infeasible for proprietary state-of-the-art detectors, and (ii) generate image-specific attacks through expensive online optimization. To address these limitations, we propose PolyJuice, the first black-box, image-agnostic red-teaming method for SIDs, based on an observed distribution shift in the T2I latent space between samples correctly and incorrectly classified by the SID. PolyJuice generates attacks by (i) identifying the direction of this shift through a lightweight offline process that only requires black-box access to the SID, and (ii) exploiting this direction by universally steering all generated images towards the SID's failure modes. PolyJuice-steered T2I models are significantly more effective at deceiving SIDs (up to 84%) compared to their unsteered counterparts. We also show that the steering directions can be estimated efficiently at lower resolutions and transferred to higher resolutions using simple interpolation, reducing computational overhead. Finally, tuning SID models on PolyJuice-augmented datasets notably enhances the performance of the detectors (up to 30%).

Petr Grinberg, Ankur Kumar, Surya Koppisetti et al.

Evaluating explainability techniques, such as SHAP and LRP, in the context of audio deepfake detection is challenging due to lack of clear ground truth annotations. In the cases when we are able to obtain the ground truth, we find that these methods struggle to provide accurate explanations. In this work, we propose a novel data-driven approach to identify artifact regions in deepfake audio. We consider paired real and vocoded audio, and use the difference in time-frequency representation as the ground-truth explanation. The difference signal then serves as a supervision to train a diffusion model to expose the deepfake artifacts in a given vocoded audio. Experimental results on the VocV4 and LibriSeVoc datasets demonstrate that our method outperforms traditional explainability techniques, both qualitatively and quantitatively.

Trevine Oorloff, Surya Koppisetti, Nicolò Bonettini et al.

With the rapid growth in deepfake video content, we require improved and generalizable methods to detect them. Most existing detection methods either use uni-modal cues or rely on supervised training to capture the dissonance between the audio and visual modalities. While the former disregards the audio-visual correspondences entirely, the latter predominantly focuses on discerning audio-visual cues within the training corpus, thereby potentially overlooking correspondences that can help detect unseen deepfakes. We present Audio-Visual Feature Fusion (AVFF), a two-stage cross-modal learning method that explicitly captures the correspondence between the audio and visual modalities for improved deepfake detection. The first stage pursues representation learning via self-supervision on real videos to capture the intrinsic audio-visual correspondences. To extract rich cross-modal representations, we use contrastive learning and autoencoding objectives, and introduce a novel audio-visual complementary masking and feature fusion strategy. The learned representations are tuned in the second stage, where deepfake classification is pursued via supervised learning on both real and fake videos. Extensive experiments and analysis suggest that our novel representation learning paradigm is highly discriminative in nature. We report 98.6% accuracy and 99.1% AUC on the FakeAVCeleb dataset, outperforming the current audio-visual state-of-the-art by 14.9% and 9.9%, respectively.

Miao Zhang, Zee fryer, Ben Colman et al.

Machine learning model bias can arise from dataset composition: correlated sensitive features can distort the downstream classification model's decision boundary and lead to performance differences along these features. Existing de-biasing works tackle the most prominent bias features, such as colors of digits or background of animals. However, real-world datasets often include a large number of feature correlations that intrinsically manifest in the data as common sense information. Such spurious visual cues can further reduce model robustness. Thus, domain practitioners desire a comprehensive understanding of correlations and the flexibility to address relevant biases. To this end, we propose a novel framework to extract comprehensive biases in image datasets based on textual descriptions, a common sense-rich modality. Specifically, features are constructed by clustering noun phrase embeddings with similar semantics. The presence of each feature across the dataset is inferred, and their co-occurrence statistics are measured, with spurious correlations optionally examined by a human-in-the-loop module. Downstream experiments show that our method uncovers novel model biases in multiple image benchmark datasets. Furthermore, the discovered bias can be mitigated by simple data re-weighting to de-correlate the features, outperforming state-of-the-art unsupervised bias mitigation methods.

Yang Gao, Tyler Vuong, Mahsa Elyasi et al.

State-of-the-art methods for audio generation suffer from fingerprint artifacts and repeated inconsistencies across temporal and spectral domains. Such artifacts could be well captured by the frequency domain analysis over the spectrogram. Thus, we propose a novel use of long-range spectro-temporal modulation feature -- 2D DCT over log-Mel spectrogram for the audio deepfake detection. We show that this feature works better than log-Mel spectrogram, CQCC, MFCC, as a suitable candidate to capture such artifacts. We employ spectrum augmentation and feature normalization to decrease overfitting and bridge the gap between training and test dataset along with this novel feature introduction. We developed a CNN-based baseline that achieved a 0.0849 t-DCF and outperformed the previously top single systems reported in the ASVspoof 2019 challenge. Finally, by combining our baseline with our proposed 2D DCT spectro-temporal feature, we decrease the t-DCF score down by 14% to 0.0737, making it a state-of-the-art system for spoofing detection. Furthermore, we evaluate our model using two external datasets, showing the proposed feature's generalization ability. We also provide analysis and ablation studies for our proposed feature and results.

Eric Engelhart, Mahsa Elyasi, Gaurav Bharaj

Grapheme-to-Phoneme (G2P) models convert words to their phonetic pronunciations. Classic G2P methods include rule-based systems and pronunciation dictionaries, while modern G2P systems incorporate learning, such as, LSTM and Transformer-based attention models. Usually, dictionary-based methods require significant manual effort to build, and have limited adaptivity on unseen words. And transformer-based models require significant training data, and do not generalize well, especially for dialects with limited data. We propose a novel use of transformer-based attention model that can adapt to unseen dialects of English language, while using a small dictionary. We show that our method has potential applications for accent transfer for text-to-speech, and for building robust G2P models for dialects with limited pronunciation dictionary size. We experiment with two English dialects: Indian and British. A model trained from scratch using 1000 words from British English dictionary, with 14211 words held out, leads to phoneme error rate (PER) of 26.877%, on a test set generated using the full dictionary. The same model pretrained on CMUDict American English dictionary, and fine-tuned on the same dataset leads to PER of 2.469% on the test set.

David Ferman, Gaurav Bharaj

We propose a general purpose approach to detect landmarks with improved temporal consistency, and personalization. Most sparse landmark detection methods rely on laborious, manually labelled landmarks, where inconsistency in annotations over a temporal volume leads to sub-optimal landmark learning. Further, high-quality landmarks with personalization is often hard to achieve. We pose landmark detection as an image translation problem. We capture two sets of unpaired marked (with paint) and unmarked videos. We then use a generative adversarial network and cyclic consistency to predict deformations of landmark templates that simulate markers on unmarked images until these images are indistinguishable from ground-truth marked images. Our novel method does not rely on manually labelled priors, is temporally consistent, and image class agnostic -- face, and hand landmarks detection examples are shown.

Mahsa Elyasi, Gaurav Bharaj

Neural sequence-to-sequence text-to-speech synthesis (TTS), such as Tacotron-2, transforms text into high-quality speech. However, generating speech with natural prosody still remains a challenge. Yasuda et. al. show that unlike natural speech, Tacotron-2's encoder doesn't fully represent prosodic features (e.g. syllable stress in English) from characters, and result in flat fundamental frequency variations. In this work, we propose a novel carefully designed strategy for conditioning Tacotron-2 on two fundamental prosodic features in English -- stress syllable and pitch accent, that help achieve more natural prosody. To this end, we use of a classifier to learn these features in an end-to-end fashion, and apply feature conditioning at three parts of Tacotron-2's Text-To-Mel Spectrogram: pre-encoder, post-encoder, and intra-decoder. Further, we show that jointly conditioned features at pre-encoder and intra-decoder stages result in prosodically natural synthesized speech (vs. Tacotron-2), and allows the model to produce speech with more accurate pitch accent and stress patterns. Quantitative evaluations show that our formulation achieves higher fundamental frequency contour correlation, and lower Mel Cepstral Distortion measure between synthesized and natural speech. And subjective evaluation shows that the proposed method's Mean Opinion Score of 4.14 fairs higher than baseline Tacotron-2, 3.91, when compared against natural speech (LJSpeech corpus), 4.28.

Abdallah Dib, Gaurav Bharaj, Junghyun Ahn et al.

We present a differentiable ray-tracing based novel face reconstruction approach where scene attributes - 3D geometry, reflectance (diffuse, specular and roughness), pose, camera parameters, and scene illumination - are estimated from unconstrained monocular images. The proposed method models scene illumination via a novel, parameterized virtual light stage, which in-conjunction with differentiable ray-tracing, introduces a coarse-to-fine optimization formulation for face reconstruction. Our method can not only handle unconstrained illumination and self-shadows conditions, but also estimates diffuse and specular albedos. To estimate the face attributes consistently and with practical semantics, a two-stage optimization strategy systematically uses a subset of parametric attributes, where subsequent attribute estimations factor those previously estimated. For example, self-shadows estimated during the first stage, later prevent its baking into the personalized diffuse and specular albedos in the second stage. We show the efficacy of our approach in several real-world scenarios, where face attributes can be estimated even under extreme illumination conditions. Ablation studies, analyses and comparisons against several recent state-of-the-art methods show improved accuracy and versatility of our approach. With consistent face attributes reconstruction, our method leads to several style -- illumination, albedo, self-shadow -- edit and transfer applications, as discussed in the paper.

Ayush Tewari, Mohamed Elgharib, Gaurav Bharaj et al.

StyleGAN generates photorealistic portrait images of faces with eyes, teeth, hair and context (neck, shoulders, background), but lacks a rig-like control over semantic face parameters that are interpretable in 3D, such as face pose, expressions, and scene illumination. Three-dimensional morphable face models (3DMMs) on the other hand offer control over the semantic parameters, but lack photorealism when rendered and only model the face interior, not other parts of a portrait image (hair, mouth interior, background). We present the first method to provide a face rig-like control over a pretrained and fixed StyleGAN via a 3DMM. A new rigging network, RigNet is trained between the 3DMM's semantic parameters and StyleGAN's input. The network is trained in a self-supervised manner, without the need for manual annotations. At test time, our method generates portrait images with the photorealism of StyleGAN and provides explicit control over the 3D semantic parameters of the face.

Abdallah Dib, Gaurav Bharaj, Junghyun Ahn et al.

We present a novel strategy to automatically reconstruct 3D faces from monocular images with explicitly disentangled facial geometry (pose, identity and expression), reflectance (diffuse and specular albedo), and self-shadows. The scene lights are modeled as a virtual light stage with pre-oriented area lights used in conjunction with differentiable Monte-Carlo ray tracing to optimize the scene and face parameters. With correctly disentangled self-shadows and specular reflection parameters, we can not only obtain robust facial geometry reconstruction, but also gain explicit control over these parameters, with several practical applications. We can change facial expressions with accurate resultant self-shadows or relight the scene and obtain accurate specular reflection and several other parameter combinations.

Ayush Tewari, Florian Bernard, Pablo Garrido et al.

Monocular image-based 3D reconstruction of faces is a long-standing problem in computer vision. Since image data is a 2D projection of a 3D face, the resulting depth ambiguity makes the problem ill-posed. Most existing methods rely on data-driven priors that are built from limited 3D face scans. In contrast, we propose multi-frame video-based self-supervised training of a deep network that (i) learns a face identity model both in shape and appearance while (ii) jointly learning to reconstruct 3D faces. Our face model is learned using only corpora of in-the-wild video clips collected from the Internet. This virtually endless source of training data enables learning of a highly general 3D face model. In order to achieve this, we propose a novel multi-frame consistency loss that ensures consistent shape and appearance across multiple frames of a subject's face, thus minimizing depth ambiguity. At test time we can use an arbitrary number of frames, so that we can perform both monocular as well as multi-frame reconstruction.