This paper presents one of the first studies on estimating group emotional states in immersive communication environments. We designed a two-phase experiment to assess group emotion recognition in both controlled and natural settings. In the forced-emotion phase, 32 participants mimicked predefined emotions in 2D and 360° video. In the natural-emotion phase, the OpenLAV dataset elicited stronger and more diverse affective responses than LIRIS-ACCEDE, indicating better suitability for triggering emotions. We also found that neutral states are mostly detected as slightly negative valence, suggesting a systematic bias in affect estimation. The study also found greater variability in valence than in arousal, potentially due to the professional setting of the experiment. Overall, the findings demonstrate the feasibility of coarse group emotion detection in 360° video showing a high correlation with participants’ self-perceptions. We believe this tool could be useful for making immersive communication more effective for audience monitoring, enhancing presentation skills, but also more inclusive for people with emotion understanding difficulties.

Abstract The deployment of Vision Transformers (ViTs) in safety-critical domains needs a clear understanding of their resilience to soft errors, since their specific layer-level vulnerabilities are currently insufficiently characterized. This work presents a dependability analysis of the ViT-Base architecture against injection-induced soft errors. Using a high-fidelity, software-level fault injection methodology with custom CUDA kernels, the study injects random bit-flips directly into the IEEE 754 binary32 floating-point representation of the intermediate data tensors resulting from the Transformer modules to quantify model accuracy degradation across increasing bit error rates. As a primary result, a vulnerability map across ViT layers is presented, confirming that the results of normalization and fully connected layers exhibit critical sensitivity to soft errors. To address these vulnerabilities, the work evaluates targeted hardening strategies. These include Fault-Aware Training (FAT), applied both globally and selectively to linear layers, as well as practical runtime mitigations such as range-based value clipping and filtering of non-numeric values. The findings demonstrate that these software-only approaches can significantly protect model accuracy.

Abstract Advances in object tracking and acoustic beamforming are driving new capabilities in surveillance, human-computer interaction, and robotics. This work presents an embedded system that integrates deep learning–based tracking with beamforming to achieve precise sound source localization and directional audio capture in dynamic environments. The approach combines single-camera depth estimation and stereo vision to enable accurate 3D localization of moving objects. A planar concentric circular microphone array constructed with MEMS microphones provides a compact, energy-efficient platform supporting 2D beam steering across azimuth and elevation. Real-time tracking outputs continuously adapt the array’s focus, synchronizing the acoustic response with the target’s position. By uniting learned spatial awareness with dynamic steering, the system maintains robust performance in the presence of multiple or moving sources. Experimental evaluation demonstrates significant gains in signal-to-interference ratio, making the design well-suited for teleconferencing, smart home devices, and assistive technologies.

Abstract Many current embedded systems comprise heterogeneous computing components including quite powerful GPUs, which enables their application across diverse sectors. This study demonstrates the efficient execution of a medium-sized self-supervised audio spectrogram transformer (SSAST) model on a low-power system-on-chip (SoC). Through comprehensive evaluation, including real time inference scenarios, we show that GPUs outperform multi-core CPUs in inference processes. Optimization techniques such as adjusting batch size, model compilation with TensorRT, and reducing data precision significantly enhance inference time, energy consumption, and memory usage. In particular, negligible accuracy degradation is observed, with post-training quantization to 8-bit integers showing less than 1% loss. This research underscores the feasibility of deploying transformer neural networks on low-power embedded devices, ensuring efficiency in time, energy, and memory, while maintaining the accuracy of the results.

Abstract Sound classification using neural networks has recently produced very accurate results. A large number of different applications use this type of sound classifiers such as controlling and monitoring the type of activity in a city or identifying different types of animals in natural environments. While traditional acoustic processing applications have been developed on high-performance computing platforms equipped with expensive multi-channel audio interfaces, the Internet of Things (IoT) paradigm requires the use of more flexible and energy-efficient systems. Although software-based platforms exist for implementing general-purpose neural networks, they are not optimized for sound classification, wasting energy and computational resources. In this work, we have used FPGAs to develop an ad hoc system where only the hardware needed for our application is synthesized, resulting in faster and more energy-efficient circuits. The results show that our developments are accelerated by a factor of 35 compared to a software-based implementation on a Raspberry Pi.

Abstract Achieving maximum parallel performance on multi-core CPUs and many-core GPUs is a challenging task depending on multiple factors. These include, for example, the number and granularity of the computations or the use of the memories of the devices. In this paper, we assess those factors by evaluating and comparing different parallelizations of the same problem on a multiprocessor containing a CPU with 40 cores and four P100 GPUs with Pascal architecture. We use, as study case, the convolutional operation behind a non-standard finite element mesh truncation technique in the context of open region electromagnetic wave propagation problems. A total of six parallel algorithms implemented using OpenMP and CUDA have been used to carry out the comparison by leveraging the same levels of parallelism on both types of platforms. Three of the algorithms are presented for the first time in this paper, including a multi-GPU method, and two others are improved versions of algorithms previously developed by some of the authors. This paper presents a thorough experimental evaluation of the parallel algorithms on a radar cross-sectional prediction problem. Results show that performance obtained on the GPU clearly overcomes those obtained in the CPU, much more so if we use multiple GPUs to distribute both data and computations. Accelerations close to 30 have been obtained on the CPU, while with the multi-GPU version accelerations larger than 250 have been achieved.