Search

Human speech conveys prosody, linguistic content, and speaker identity. This article investigates a novel speaker anonymization approach using an end-to-end network based on a Vector-Quantized Variational Auto-Encoder (VQ-VAE) to deal with these speech components. This approach is designed to disentangle these components to specifically target and modify the speaker identity while preserving the linguistic and emotionalcontent. To do so, three separate branches compute embeddings for content, prosody, and speaker identity respectively. During synthesis, taking these embeddings, the decoder of the proposed architecture is conditioned on both speaker and prosody information, allowing for capturing more nuanced emotional states and precise adjustments to speaker identification. Findings indicate that this method outperforms most baseline techniques in preserving emotional information. However, it exhibits more limited performance on other voice privacy tasks, emphasizing the need for further improvements.

Human Activity Recognition (HAR) is one of the central problems in fields such as healthcare, elderly care, and security at home. However, traditional HAR approaches face challenges including data scarcity, difficulties in model generalization, and the complexity of recognizing activities in multi-person scenarios. This paper proposes a system framework called LAHAR, based on large language models. Utilizing prompt engineering techniques, LAHAR addresses HAR in multi-person scenarios by enabling subject separation and action-level descriptions of events occurring in the environment. We validated our approach on the ARAS dataset, and the results demonstrate that LAHAR achieves comparable accuracy to the state-of-the-art method at higher resolutions and maintains robustness in multi-person scenarios.

This paper presents two models to address the problem of multi-person activity recognition using ambient sensors in a home. The first model, Seq2Res, uses a sequence generation approach to separate sensor events from different residents. The second model, BiGRU+Q2L, uses a Query2Label multi-label classifier to predict multiple activities simultaneously. Performances of these models are compared to a state-of-the-art model in different experimental scenarios, using a state-of-the-art dataset of two residents in a home instrumented with ambient sensors. These results lead to a discussion on the advantages and drawbacks of resident separation and multi-label classification for multi-person activity recognition., Comment: Context and Activity Modeling and Recognition (CoMoReA) Workshop at IEEE International Conference on Pervasive Computing and Communications (PerCom 2024), Mar 2024, Biarritz, France

This paper explores privacy-compliant group-level emotion recognition ''in-the-wild'' within the EmotiW Challenge 2023. Group-level emotion recognition can be useful in many fields including social robotics, conversational agents, e-coaching and learning analytics. This research imposes itself using only global features avoiding individual ones, i.e. all features that can be used to identify or track people in videos (facial landmarks, body poses, audio diarization, etc.). The proposed multimodal model is composed of a video and an audio branches with a cross-attention between modalities. The video branch is based on a fine-tuned ViT architecture. The audio branch extracts Mel-spectrograms and feed them through CNN blocks into a transformer encoder. Our training paradigm includes a generated synthetic dataset to increase the sensitivity of our model on facial expression within the image in a data-driven way. The extensive experiments show the significance of our methodology. Our privacy-compliant proposal performs fairly on the EmotiW challenge, with 79.24% and 75.13% of accuracy respectively on validation and test set for the best models. Noticeably, our findings highlight that it is possible to reach this accuracy level with privacy-compliant features using only 5 frames uniformly distributed on the video.

Animating still face images with deep generative models using a speech input signal is an active research topic and has seen important recent progress. However, much of the effort has been put into lip syncing and rendering quality while the generation of natural head motion, let alone the audio-visual correlation between head motion and speech, has often been neglected. In this work, we propose a multi-scale audio-visual synchrony loss and a multi-scale autoregressive GAN to better handle short and long-term correlation between speech and the dynamics of the head and lips. In particular, we train a stack of syncer models on multimodal input pyramids and use these models as guidance in a multi-scale generator network to produce audio-aligned motion unfolding over diverse time scales. Our generator operates in the facial landmark domain, which is a standard low-dimensional head representation. The experiments show significant improvements over the state of the art in head motion dynamics quality and in multi-scale audio-visual synchrony both in the landmark domain and in the image domain.

The transition angles are defined to describe the vowel-to-vowel transitions in the acoustic space of the Spectral Subband Centroids, and the findings show that they are similar among speakers and speaking rates. In this paper, we propose to investigate the usage of polar coordinates in favor of angles to describe a speech signal by characterizing its acoustic trajectory and using them in Automatic Speech Recognition. According to the experimental results evaluated on the BRAF100 dataset, the polar coordinates achieved significantly higher accuracy than the angles in the mixed and cross-gender speech recognitions, demonstrating that these representations are superior at defining the acoustic trajectory of the speech signal. Furthermore, the accuracy was significantly improved when they were utilized with their first and second-order derivatives ($\Delta$, $\Delta$$\Delta$), especially in cross-female recognition. However, the results showed they were not much more gender-independent than the conventional Mel-frequency Cepstral Coefficients (MFCCs).

In this work, we address the task of unconditional head motion generation to animate still human faces in a low-dimensional semantic space from a single reference pose. Different from traditional audio-conditioned talking head generation that seldom puts emphasis on realistic head motions, we devise a GAN-based architecture that learns to synthesize rich head motion sequences over long duration while maintaining low error accumulation levels.In particular, the autoregressive generation of incremental outputs ensures smooth trajectories, while a multi-scale discriminator on input pairs drives generation toward better handling of high- and low-frequency signals and less mode collapse.We experimentally demonstrate the relevance of the proposed method and show its superiority compared to models that attained state-of-the-art performances on similar tasks.

In this paper, we describe a graph-based algorithm that uses the features obtained by a self-supervised transformer to detect and segment salient objects in images and videos. With this approach, the image patches that compose an image or video are organised into a fully connected graph, where the edge between each pair of patches is labeled with a similarity score between patches using features learned by the transformer. Detection and segmentation of salient objects is then formulated as a graph-cut problem and solved using the classical Normalized Cut algorithm. Despite the simplicity of this approach, it achieves state-of-the-art results on several common image and video detection and segmentation tasks. For unsupervised object discovery, this approach outperforms the competing approaches by a margin of 6.1%, 5.7%, and 2.6%, respectively, when tested with the VOC07, VOC12, and COCO20K datasets. For the unsupervised saliency detection task in images, this method improves the score for Intersection over Union (IoU) by 4.4%, 5.6% and 5.2%. When tested with the ECSSD, DUTS, and DUT-OMRON datasets, respectively, compared to current state-of-the-art techniques. This method also achieves competitive results for unsupervised video object segmentation tasks with the DAVIS, SegTV2, and FBMS datasets., Comment: arXiv admin note: text overlap with arXiv:2202.11539

Transformers trained with self-supervised learning using self-distillation loss (DINO) have been shown to produce attention maps that highlight salient foreground objects. In this paper, we demonstrate a graph-based approach that uses the self-supervised transformer features to discover an object from an image. Visual tokens are viewed as nodes in a weighted graph with edges representing a connectivity score based on the similarity of tokens. Foreground objects can then be segmented using a normalized graph-cut to group self-similar regions. We solve the graph-cut problem using spectral clustering with generalized eigen-decomposition and show that the second smallest eigenvector provides a cutting solution since its absolute value indicates the likelihood that a token belongs to a foreground object. Despite its simplicity, this approach significantly boosts the performance of unsupervised object discovery: we improve over the recent state of the art LOST by a margin of 6.9%, 8.1%, and 8.1% respectively on the VOC07, VOC12, and COCO20K. The performance can be further improved by adding a second stage class-agnostic detector (CAD). Our proposed method can be easily extended to unsupervised saliency detection and weakly supervised object detection. For unsupervised saliency detection, we improve IoU for 4.9%, 5.2%, 12.9% on ECSSD, DUTS, DUT-OMRON respectively compared to previous state of the art. For weakly supervised object detection, we achieve competitive performance on CUB and ImageNet.

Urban autonomous driving in the presence of pedestrians as vulnerable road users is still a challenging and less examined research problem. This work formulates navigation in urban environments as a multi objective reinforcement learning problem. A deep learning variant of thresholded lexicographic Q-learning is presented for autonomous navigation amongst pedestrians. The multi objective DQN agent is trained on a custom urban environment developed in CARLA simulator. The proposed method is evaluated by comparing it with a single objective DQN variant on known and unknown environments. Evaluation results show that the proposed method outperforms the single objective DQN variant with respect to all aspects.

Prediction of human actions in social interactions has important applications in the design of social robots or artificial avatars. In this paper, we focus on a unimodal representation of interactions and propose to tackle interaction generation in a data-driven fashion. In particular, we model human interaction generation as a discrete multi-sequence generation problem and present SocialInteractionGAN, a novel adversarial architecture for conditional interaction generation. Our model builds on a recurrent encoder-decoder generator network and a dual-stream discriminator, that jointly evaluates the realism of interactions and individual action sequences and operates at different time scales. Crucially, contextual information on interacting participants is shared among agents and reinjected in both the generation and the discriminator evaluation processes. Experiments show that albeit dealing with low dimensional data, SocialInteractionGAN succeeds in producing high realism action sequences of interacting people, comparing favorably to a diversity of recurrent and convolutional discriminator baselines, and we argue that this work will constitute a first stone towards higher dimensional and multimodal interaction generation. Evaluations are conducted using classical GAN metrics, that we specifically adapt for discrete sequential data. Our model is shown to properly learn the dynamics of interaction sequences, while exploiting the full range of available actions., Comment: IEEE Transactions on Affective Computing, Institute of Electrical and Electronics Engineers, 2022

Decision making for autonomous driving in urban environments is challenging due to the complexity of the road structure and the uncertainty in the behavior of diverse road users. Traditional methods consist of manually designed rules as the driving policy, which require expert domain knowledge, are difficult to generalize and might give sub-optimal results as the environment gets complex. Whereas, using reinforcement learning, optimal driving policy could be learned and improved automatically through several interactions with the environment. However, current research in the field of reinforcement learning for autonomous driving is mainly focused on highway setup with little to no emphasis on urban environments. In this work, a deep reinforcement learning based decision-making approach for high-level driving behavior is proposed for urban environments in the presence of pedestrians. For this, the use of Deep Recurrent Q-Network (DRQN) is explored, a method combining state-of-the art Deep Q-Network (DQN) with a long term short term memory (LSTM) layer helping the agent gain a memory of the environment. A 3-D state representation is designed as the input combined with a well defined reward function to train the agent for learning an appropriate behavior policy in a real-world like urban simulator. The proposed method is evaluated for dense urban scenarios and compared with a rule-based approach and results show that the proposed DRQN based driving behavior decision maker outperforms the rule-based approach.

This article presents our unimodal privacy-safe and non-individual proposal for the audio-video group emotion recognition subtask at the Emotion Recognition in the Wild (EmotiW) Challenge 2020 1. This sub challenge aims to classify in the wild videos into three categories: Positive, Neutral and Negative. Recent deep learning models have shown tremendous advances in analyzing interactions between people, predicting human behavior and affective evaluation. Nonetheless, their performance comes from individual-based analysis, which means summing up and averaging scores from individual detections, which inevitably leads to some privacy issues. In this research, we investigated a frugal approach towards a model able to capture the global moods from the whole image without using face or pose detection, or any individual-based feature as input. The proposed methodology mixes state-of-the-art and dedicated synthetic corpora as training sources. With an in-depth exploration of neural network architectures for group-level emotion recognition, we built a VGG-based model achieving 59.13% accuracy on the VGAF test set (eleventh place of the challenge). Given that the analysis is unimodal based only on global features and that the performance is evaluated on a real-world dataset, these results are promising and let us envision extending this model to multimodality for classroom ambiance evaluation, our final target application.

This article reports on an investigation of the use of convolutional neural networks to predict the visual attention of chess players. The visual attention model described in this article has been created to generate saliency maps that capture hierarchical and spatial features of chessboard, in order to predict the probability fixation for individual pixels Using a skip-layer architecture of an autoencoder, with a unified decoder, we are able to use multiscale features to predict saliency of part of the board at different scales, showing multiple relations between pieces. We have used scan path and fixation data from players engaged in solving chess problems, to compute 6600 saliency maps associated to the corresponding chess piece configurations. This corpus is completed with synthetically generated data from actual games gathered from an online chess platform. Experiments realized using both scan-paths from chess players and the CAT2000 saliency dataset of natural images, highlights several results. Deep features, pretrained on natural images, were found to be helpful in training visual attention prediction for chess. The proposed neural network architecture is able to generate meaningful saliency maps on unseen chess configurations with good scores on standard metrics. This work provides a baseline for future work on visual attention prediction in similar contexts.

In this paper we present results from recent experiments that suggest that chess players associate emotions to game situations and reactively use these associations to guide search for planning and problem solving. We describe the design of an instrument for capturing and interpreting multimodal signals of humans engaged in solving challenging problems. We review results from a pilot experiment with human experts engaged in solving challenging problems in Chess that revealed an unexpected observation of rapid changes in emotion as players attempt to solve challenging problems. We propose a cognitive model that describes the process by which subjects select chess chunks for use in interpretation of the game situation and describe initial results from a second experiment designed to test this model.

Autonomous Vehicles navigating in urban areas have a need to understand and predict future pedestrian behavior for safer navigation. This high level of situational awareness requires observing pedestrian behavior and extrapolating their positions to know future positions. While some work has been done in this field using Hidden Markov Models (HMMs), one of the few observed drawbacks of the method is the need for informed priors for learning behavior. In this work, an extension to the Growing Hidden Markov Model (GHMM) method is proposed to solve some of these drawbacks. This is achieved by building on existing work using potential cost maps and the principle of Natural Vision. As a consequence, the proposed model is able to predict pedestrian positions more precisely over a longer horizon compared to the state of the art. The method is tested over "legal" and "illegal" behavior of pedestrians, having trained the model with sparse observations and partial trajectories. The method, with no training data, is compared against a trained state of the art model. It is observed that the proposed method is robust even in new, previously unseen areas., Comment: 15 th International Conference on Control, Automation, Robotics and Vision (ICARCV 2018), Nov 2018, Singapore, Singapore

In a multiuser context, the Bluetooth data from the smartphone could give an approximation of the distance between users. Meanwhile, the Wi-Fi data can be used to calculate the user's position directly. However, both the Wi-Fi-based position outputs and Bluetooth-based distances are affected by some degree of noise. In our work, we propose several approaches to combine the two types of outputs for improving the tracking accuracy in the context of collaborative positioning. The two proposed approaches attempt to build a model for measuring the errors of the Bluetooth output and Wi-Fi output. In a non-temporal approach, the model establishes the relationship in a specific interval of the Bluetooth output and Wi-Fi output. In a temporal approach, the error measurement model is expanded to include the time component between users' movement. To evaluate the performance of the two approaches, we collected the data from several multiuser scenarios in indoor environment. The results show that the proposed approaches could reach a distance error around 3.0m for 75 percent of time, which outperforms the positioning results of the standard Wi-Fi fingerprinting model., Comment: International Conference on Indoor Positioning and Indoor Navigation (IPIN), Sep 2018, Nantes, France

This article deals with the specific context of an autonomous car navigating in an urban center within a shared space between pedestrians and cars. The driver delegates the control to the autonomous system while remaining seated in the driver's seat. The proposed study aims at giving a first insight into the definition of human perception of space applied to vehicles by testing the existence of a personal space around the car.It aims at measuring proxemic information about the driver's comfort zone in such conditions.Proxemics, or human perception of space, has been largely explored when applied to humans or to robots, leading to the concept of personal space, but poorly when applied to vehicles. In this article, we highlight the existence and the characteristics of a zone of comfort around the car which is not correlated to the risk of a collision between the car and other road users. Our experiment includes 19 volunteers using a virtual reality headset to look at 30 scenarios filmed in 360{\textdegree} from the point of view of a passenger sitting in the driver's seat of an autonomous car.They were asked to say "stop" when they felt discomfort visualizing the scenarios.As said, the scenarios voluntarily avoid collision effect as we do not want to measure fear but discomfort.The scenarios involve one or three pedestrians walking past the car at different distances from the wings of the car, relative to the direction of motion of the car, on both sides. The car is either static or moving straight forward at different speeds.The results indicate the existence of a comfort zone around the car in which intrusion causes discomfort.The size of the comfort zone is sensitive neither to the side of the car where the pedestrian passes nor to the number of pedestrians. In contrast, the feeling of discomfort is relative to the car's motion (static or moving).Another outcome from this study is an illustration of the usage of first person 360{\textdegree} video and a virtual reality headset to evaluate feelings of a passenger within an autonomous car.

In this paper we present the first results of a pilot experiment in the capture and interpretation of multimodal signals of human experts engaged in solving challenging chess problems. Our goal is to investigate the extent to which observations of eye-gaze, posture, emotion and other physiological signals can be used to model the cognitive state of subjects, and to explore the integration of multiple sensor modalities to improve the reliability of detection of human displays of awareness and emotion. We observed chess players engaged in problems of increasing difficulty while recording their behavior. Such recordings can be used to estimate a participant's awareness of the current situation and to predict ability to respond effectively to challenging situations. Results show that a multimodal approach is more accurate than a unimodal one. By combining body posture, visual attention and emotion, the multimodal approach can reach up to 93% of accuracy when determining player's chess expertise while unimodal approach reaches 86%. Finally this experiment validates the use of our equipment as a general and reproducible tool for the study of participants engaged in screen-based interaction and/or problem solving.

Recognition of intentions is a subconscious cognitive process vital to human communication. This skill enables anticipation and increases the quality of interactions between humans. Within the context of engagement, non-verbal signals are used to communicate the intention of starting the interaction with a partner. In this paper, we investigated methods to detect these signals in order to allow a robot to know when it is about to be addressed. Originality of our approach resides in taking inspiration from social and cognitive sciences to perform our perception task. We investigate meaningful features, i.e. human readable features, and elicit which of these are important for recognizing someone's intention of starting an interaction. Classically, spatial information like the human position and speed, the human-robot distance are used to detect the engagement. Our approach integrates multimodal features gathered using a companion robot equipped with a Kinect. The evaluation on our corpus collected in spontaneous conditions highlights its robustness and validates the use of such a technique in a real environment. Experimental validation shows that multimodal features set gives better precision and recall than using only spatial and speed features. We also demonstrate that 7 selected features are sufficient to provide a good starting engagement detection score. In our last investigation, we show that among our full 99 features set, the space reduction is not a solved task. This result opens new researches perspectives on multimodal engagement detection.