The information published in newspapers reflects the opinions and viewpoints that society holds on the principal public issues. Some topics can be particularly sensitive to gender biases, such as technology, politics, or sports, where, despite the growing relevance of women, the media tends to focus on men as protagonists and sources of reference. When women do appear, they are not treated equally. Analyzing how the information is presented from a gender perspective is necessary to assess progress toward equality in the journalism sector and society at large. These analyses are very costly to produce due to the time required and the shortage of experts in the field, so they are not as frequent as would be ideal. In this paper, we introduce IRIS, a tool that helps experts identify gender biases in news publications. The tool is designed to process content qualitatively and quantitatively, using variables defined in collaboration with researchers with extensive experience in gender studies. This first version of IRIS aims to capture human expertise and automatically generate a basic report to present analysis results from a gender perspective.

Soundscapes are a fundamental component of the context of real-world scenarios necessary to provide situation awareness in cyberphysical systems. However, real life conditions challenge the achievements of most artificial intelligence based methods to characterize them. In this paper we have identified the challenges for Acoustic Scene Classification (ASC) in daily-life scenarios of Gender-based Violence Victims (GBVV) with the aim of providing context information for the multimodal detection of violent scenarios. Specifically, these entail data imbalance, low quality of recordings, low quality of labels, inter-subject variability, low inference time and robustness requirements to various changing conditions. The unique Women and Emotion in real LIfe affectiVE computing dataset (WELIVE) comprising unconstrained, user-centered, in-the-wild daily-life recordings of 14 gender-based violence victims for 7-10 days with self-reported semantic location labels is used to evaluate hand-crafted and learnt features using pre-trained models transferred from existing acoustic event classifiers. We propose a distilled version of the Large-Scale Pretrained Audio Neural Networks for Audio Pattern Recognition (PANNs) trained on AudioSet and adapted to the ASC domain, achieving the best trade-off for the challenges identified. Our method demonstrates a high information transfer factor of 1.843 with a fast inference time of 1 . 989 ± 0 . 052 milliseconds per 24-sample batch, highlighting its potential for real-time, context-aware applications in sensitive, unconstrained environments.

Technologies developed at universities are direct means to conceive a teaching plan for promoting science, technology, engineering and mathematics (STEM) at the K-12 level of education. This work-in-progress develops this notion: We elaborate a teaching plan for K-12 students based on technologies for recording plant bioelectrical activity. The teaching plan sketches hands-on activities, where pupils can self-assemble the electronic components such as the electro-potential sensor, analog-to-digital (ADC) converter, and Arduino board for processing, as designed by research students at Politecnico di Milano. Using our project as a blueprint, we aim to support other educators at universities to promote further STEM and expose pupils to technical developments at universities.

Reverse-engineering a living creature’s functional connectome is a key challenge in neuroscience. Using K-Formal Concept Analysis, we explore the olfactory perception of Caenorhabditis elegans uncovering robust multimodal relationships between odorants and sensory neurons. Our findings reveal a many-to-many encoding scheme, where neurons respond to multiple odorants, and each odorant activates several neurons—an efficient strategy for sensory encoding in compact connectomes. Robustness tests confirm the stability of our results across dilutions and ablations.Overall, our work highlights the potential of K-Formal Concept Analysis and related techniques to reverse-engineer the sensory perception phase of the sensation-action loop in model organisms, paving the way for a deeper understanding of the energy-efficient intelligence of C. elegans.

Understanding soundscapes is essential for making sense of real-world scenarios in complex environments. However, real-life conditions present significant challenges for AI-based methods, particularly due to the highly imbalanced nature of Audio Tagging problems. In this work, we investigate the impact of data imbalance on the training dynamics of state-of-the-art models for Acoustic Scene Classification. Using the DCASE TAU Urban Acoustic Scenes 2022 dataset and the CP-Mobile model, we introduce controlled imbalance scenarios and analyze their effect through the Entropic Triangle framework. Our findings reveal that the training dynamics are strongly influenced by the chosen balancing approach. It also suggests longer training periods with conventional optimizer for re-balanced classification metrics. In this way, this study provides new insights into the role of entropy-based analysis in developing robust Acoustic Scene Classification systems for real-world applications. © 2025 International Speech Communication Association. All rights reserved.

The main research motivation of this article is the fight against gender-based violence and achieving gender equality from a technological perspective. The solution proposed in this work goes beyond currently existing panic buttons, needing to be manually operated by the victims under difficult circumstances. Instead, Bindi, our end-to-end autonomous multimodal system, relies on artificial intelligence methods to automatically identify violent situations, based on detecting fear-related emotions, and trigger a protection protocol, if necessary. To this end, Bindi integrates modern state-of-the-art technologies, such as the Internet of Bodies, affective computing, and cyber–physical systems, leveraging: 1) affective Internet of Things (IoT) with auditory and physiological commercial off-the-shelf smart sensors embedded in wearable devices; 2) hierarchical multisensorial information fusion; and 3) the edge-fog-cloud IoT architecture. This solution is evaluated using our own data set named WEMAC, a very recently collected and freely available collection of data comprising the auditory and physiological responses of 47 women to several emotions elicited by using a virtual reality environment. On this basis, this work provides an analysis of multimodal late fusion strategies to combine the physiological and speech data processing pipelines to identify the best intelligence engine strategy for Bindi. In particular, the best data fusion strategy reports an overall fear classification accuracy of 63.61% for a subject-independent approach. Both a power consumption study and an audio data processing pipeline to detect violent acoustic events complement this analysis. This research is intended as an initial multimodal baseline that facilitates further work with real-life elicited fear in women.

We introduce a framework for the evaluation of multiclass classifiers by exploring their confusion matrices. Instead of using error-counting measures of performance, we concentrate in quantifying the information transfer from true to estimated labels using information-theoretic measures. First, the Entropy Triangle allows us to visualize the balance of mutual information, variation of information, and the deviation from uniformity in the true and estimated label distributions. Next, the Entropy-Modified Accuracy allows us to rank classifiers by performance while the Normalized Information Transfer rate allows us to evaluate classifiers by the amount of information accrued during learning. Finally, if the question rises to elucidate which errors are systematically committed by the classifier, we use a generalization of Formal Concept Analysis to elicit such knowledge. All such techniques can be applied either to artificially or biologically embodied classifiers-e.g., human performance on perceptual tasks. We instantiate the framework in a number of examples to provide guidelines for the use of these tools in the case of assessing single classifiers or populations of them-whether induced with the same technique or not-either on single tasks or in a set of them. These include well-known UCI tasks and the more complex KDD cup 99 competition on Intrusion Detection.

In this paper we investigate into analogue of a Singular Value Decomposition for rectangular matrices over a complete idempotent semifields (i-SVD), many of whose instances-the complete schedule algebra or completed max-plus semifield, the tropical algebra, and the max-times algebra-are useful in a range of applications. We carry on further the task of eliciting the relation between i-SVD and the extension of Formal Concept Analysis to complete idempotent semifields ($\mathcal{K}$-FCA) started in prior work. We find out that for each relation with entries considered in a complete idempotent semifield, for each of the four types of Galois connections defined by the relation, there are two basis of left- and right-singular vectors to choose from for reconstructing the matrix. These are each of the join-dense or meet-dense sets object or attribute concepts of the concept lattice created by the connection. We suggest why these can never form an orthonormal basis, and refer to prior art describing the spectral lattices of such matrices for meaningful representations in the idempotent settings.

BACKGROUND: Gene Expression Data (GED) analysis poses a great challenge to the scientific community that can be framed into the Knowledge Discovery in Databases (KDD) and Data Mining (DM) paradigm. Biclustering has emerged as the machine learning method of choice to solve this task, but its unsupervised nature makes result assessment problematic. This is often addressed by means of Gene Set Enrichment Analysis (GSEA). RESULTS: We put forward a framework in which GED analysis is understood as an Exploratory Data Analysis (EDA) process where we provide support for continuous human interaction with data aiming at improving the step of hypothesis abduction and assessment. We focus on the adaptation to human cognition of data interpretation and visualization of the output of EDA. First, we give a proper theoretical background to bi-clustering using Lattice Theory and provide a set of analysis tools revolving around [Formula: see text]-Formal Concept Analysis ([Formula: see text]-FCA), a lattice-theoretic unsupervised learning technique for real-valued matrices. By using different kinds of cost structures to quantify expression we obtain different sequences of hierarchical bi-clusterings for gene under- and over-expression using thresholds. Consequently, we provide a method with interleaved analysis steps and visualization devices so that the sequences of lattices for a particular experiment summarize the researcher's vision of the data. This also allows us to define measures of persistence and robustness of biclusters to assess them. Second, the resulting biclusters are used to index external omics databases-for instance, Gene Ontology (GO)-thus offering a new way of accessing publicly available resources. This provides different flavors of gene set enrichment against which to assess the biclusters, by obtaining their p-values according to the terminology of those resources. We illustrate the exploration procedure on a real data example confirming results previously published. CONCLUSIONS: The GED analysis problem gets transformed into the exploration of a sequence of lattices enabling the visualization of the hierarchical structure of the biclusters with a certain degree of granularity. The ability of FCA-based bi-clustering methods to index external databases such as GO allows us to obtain a quality measure of the biclusters, to observe the evolution of a gene throughout the different biclusters it appears in, to look for relevant biclusters-by observing their genes and what their persistence is-to infer, for instance, hypotheses on their function.

A two-level variable bit rate (VBR) control algorithm for hierarchical video coding, specifically tailored for the new High Efficiency Video Coding (HEVC) standard, is presented here. A long-term level monitors the current bit count along a sliding window of a few seconds, comprising several intra-periods (IPs) and shifted on an IP basis. This long-term view allows the accommodation of the naturally occurring rate variations at a slow pace, avoiding the annoying sharp quality changes commonly appearing when non-sliding window approaches are used. The bit excesses or defects observed at this level are evenly delivered to a short-term level mechanism that establishes target bit budgets for a narrower sliding window covering a single IP and shifting on a frame basis. At this level, an adequate quantization parameter is estimated to comply with the designated target bit rate. Recommended test conditions as well as two few minutes long video sequences with scene cuts have been used for the assessment of the proposed VBR controller. Comparisons with a state-of-the-art rate control algorithm have produced good results in terms of quality consistency, in exchange for moderate rate-distortion performance losses.

In this paper, we present advances in the modeling of the masking behavior of the human auditory system (HAS) to enhance the robustness of the feature extraction stage in automatic speech recognition (ASR). The solution adopted is based on a nonlinear filtering of a spectro-temporal representation applied simultaneously to both frequency and time domains - as if it were an image - using mathematical morphology operations. A particularly important component of this architecture is the so-called structuring element (SE) that in the present contribution is designed as a single three-dimensional pattern using physiological facts, in such a way that closely resembles the masking phenomena taking place in the cochlea. A proper choice of spectro-temporal representation lends validity to the model throughout the whole frequency spectrum and intensity spans assuming the variability of the masking properties of the HAS in these two domains. The best results were achieved with the representation introduced as part of the power normalized cepstral coefficients (PNCC) together with a spectral subtraction step. This method has been tested on Aurora 2, Wall Street Journal and ISOLET databases including both classical hidden Markov model (HMM) and hybrid artificial neural networks (ANN)-HMM back-ends. In these, the proposed front-end analysis provides substantial and significant improvements compared to baseline techniques: up to 39.5% relative improvement compared to MFCC, and 18.7% compared to PNCC in the Aurora 2 database.

Proceedings of: 15th Annual Conference of the International Speech Communication Association. Singapore, September 14-18, 2014.

Perceptual coding has become of great interest in modern video coding due to the need for higher compression rates. Many previous works have been carried out to incorporate perceptual information to hybrid video encoders, either modifying the quantization parameter according to a certain perceptual resource allocation map or preprocessing video sequences for removing information that is not perceptually relevant. The first strategy is limited by the presence of blocking artifacts and the second one lacks of adaptation to video content. In this paper, a novel and simple approach is proposed, which performs a smart filtering prior to the encoding process preserving both the structural and motion information. The experiments prove that the use of proposed method implemented on an H.264 encoder significantly improves its perceptual quality for low bit rates.

A simplified protocol and associated metrics based on Signal Detection Theory (SDT) for subjective Video Quality Assessment (VQA) is proposed with the aim of filling the gap existing between the lack of discrimination abilities of objective Quality Estimates (specially when perceptually motivated processing methods are involved) and the costly normative subjective quality tests. The proposed protocol employs a reduced number of assessors and provides a quality ranking of the methods being evaluated. It is intended for providing the rapid experimental turn around necessary for developing algorithms. We have validated our proposal by corroborating with our test a well-known result for the video coding community: the quality benefits of including an in-loop deblocking filter. A software interface to design and administrate the test is also made publicly available.

In the last years, support vector machines (SVMs) have shown excellent performance in many applications, especially in the presence of noise. In particular, SVMs offer several advantages over artificial neural networks (ANNs) that have attracted the attention of the speech processing community. Nevertheless, their high computational requirements prevent them from being used in practice in automatic speech recognition (ASR), where ANNs have proven to be successful. The high complexity of SVMs in this context arises from the use of huge speech training databases with millions of samples and highly overlapped classes. This paper suggests the use of a weighted least squares (WLS) training procedure that facilitates the possibility of imposing a compact semiparametric model on the SVM, which results in a dramatic complexity reduction. Such a complexity reduction with respect to conventional SVMs, which is between two and three orders of magnitude, allows the proposed hybrid WLS-SVC/HMM system to perform real-time speech decoding on a connected-digit recognition task (SpeechDat Spanish database). The experimental evaluation of the proposed system shows encouraging performance levels in clean and noisy conditions, although further improvements are required to reach the maturity level of current context-dependent HMM-based recognizers.

Hybrid speech recognizers, where the estimation of the emission pdf of the states of hidden Markov models (HMMs), usually carried out using Gaussian mixture models (GMMs), is substituted by artificial neural networks (ANNs) have several advantages over the classical systems. However, to obtain performance improvements, the computational requirements are heavily increased because of the need to train the ANN. Departing from the observation of the remarkable skewness of speech data, this paper proposes sifting out the training set and balancing the amount of samples per class. With this method, the training time has been reduced 18 times while obtaining performances similar to or even better than those with the whole database, especially in noisy environments. However, the application of these reduced sets is not straightforward. To avoid the mismatch between training and testing conditions created by the modification of the distribution of the training data, a proper scaling of the a posteriori probabilities obtained and a resizing of the context window need to be performed as demonstrated in this paper.

The use of huge databases in ASR has become an important source of ASR system improvements in the last years. However, their use demands an increase of the computational resources necessary to train the recognizers. Several techniques have been proposed in the literature with the purpose of making a better use of these enormous databases by selecting the most ‘informative ‘ portions and thus reducing the computational burden. In this paper, we present a technique to select samples from a database that allows us to obtain similar results in MLP-based feature extraction stages by using around 60 % of the data. 1.

The Global System for Mobile (GSM) environment encompasses three main problems for automatic speech recognition (ASR) systems: noisy scenarios, source coding distortion, and transmission errors. The first one has already received much attention; however, source coding distortion and transmission errors must be explicitly addressed. In this paper, we propose an alternative front-end for speech recognition over GSM networks. This front-end is specially conceived to be effective against source coding distortion and transmission errors. Specifically, we suggest extracting the recognition feature vectors directly from the encoded speech (i.e., the bitstream) instead of decoding it and subsequently extracting the feature vectors. This approach offers two significant advantages. First, the recognition system is only affected by the quantization distortion of the spectral envelope. Thus, we are avoiding the influence of other sources of distortion as a result of the encoding-decoding process. Second, when transmission errors occur, our front-end becomes more effective since it is not affected by errors in bits allocated to the excitation signal. We have considered the half and the full-rate standard codecs and compared the proposed front-end with the conventional approach in two ASR tasks, namely, speaker-independent isolated digit recognition and speaker-independent continuous speech recognition. In general, our approach outperforms the conventional procedure, for a variety of simulated channel conditions. Furthermore, the disparity increases as the network conditions worsen.

The paper tackles the problem of noisy speech recognition. In particular, we present a novel approach to the design of filters for processing the modulation spectrum, that we call linear equalization. We postulate that, as long as the distortion of the spectral parameters due to noise can be modeled as linear, an advantageous solution consists of estimating this linear perturbation system and designing its inverse system (the equalizer). Our experimental results show that the proposed method is very effective for three of the five types of noise considered.

The Internet Protocol (IP) environment poses two relevant sources of distortion to the speech recognition problem: lossy speech coding and packet loss. In this paper, we propose a new front-end for speech recognition over IP networks. Specifically, we suggest extracting the recognition feature vectors directly from the encoded speech (i.e., the bit stream) instead of decoding it and subsequently extracting the feature vectors. This approach offers two significant benefits. First, the recognition system is only affected by the quantization distortion of the spectral envelope. Thus, we are avoiding the influence of other sources of distortion due to the encoding-decoding process. Second, when packet loss occurs, our front-end becomes more effective since it is not constrained to the error handling mechanism of the codec. We have considered the ITU G.723.1 standard codec, which is one of the most preponderant coding algorithms in voice over IP (VoIP) and compared the proposed front-end with the conventional approach in two automatic speech recognition (ASR) tasks, namely, speaker-independent isolated digit recognition and speaker-independent continuous speech recognition. In general, our approach outperforms the conventional procedure, for a variety of simulated packet loss rates. Furthermore, the improvement is higher as network conditions worsen.