arXiv subjects

Mariel Estevez

Publications and source records attributed to Mariel Estevez.

6 recordsLinked to original sources

arXiv2026

A Fair and Transparent Framework for Speech-Based Depression Detection: Balancing Interpretability and Performance

While speech provides rich, non-invasive biomarkers for mental-health assessment, clinical adoption is limited by opaque models and potential demographic bias. In this work we propose a methodological framework to evaluate robustness and interpretability for automated depression detection on the extended DAIC-WOZ dataset using low-complexity machine learning baselines (RF, SVM, and MLP) chosen to mitigate overfitting and enhance generalization in combination with human-understandable acoustic features (MFCCs, eGeMAPS). To balance accuracy with clinical trust, we leverage explainability methods (LIME and SHAP) for feature selection, validating our findings with statistical significance tests and demographic fairness analyses to mitigate spurious, artifact-driven correlations. Empirical results demonstrate that an optimized subset of explainable AI (XAI)-selected features combined with an MLP architecture achieves a state-of-the-art test accuracy of 82\%. Ultimately, this work provides a transparent framework for robust and ethical assistive technologies that can be applied to any other binary task.

eess.AS↗

arXiv2025

Study on the Fairness of Speaker Verification Systems on Underrepresented Accents in English

Speaker verification (SV) systems are currently being used to make sensitive decisions like giving access to bank accounts or deciding whether the voice of a suspect coincides with that of the perpetrator of a crime. Ensuring that these systems are fair and do not disfavor any particular group is crucial. In this work, we analyze the performance of several state-of-the-art SV systems across groups defined by the accent of the speakers when speaking English. To this end, we curated a new dataset based on the VoxCeleb corpus where we carefully selected samples from speakers with accents from different countries. We use this dataset to evaluate system performance for several SV systems trained with VoxCeleb data. We show that, while discrimination performance is reasonably robust across accent groups, calibration performance degrades dramatically on some accents that are not well represented in the training data. Finally, we show that a simple data balancing approach mitigates this undesirable bias, being particularly effective when applied to our recently-proposed discriminative condition-aware backend.

eess.AS↗

arXiv2025

Beyond Global Metrics: A Fairness Analysis for Interpretable Voice Disorder Detection Systems

We conducted a comprehensive analysis of an Automatic Voice Disorders Detection (AVDD) system using existing voice disorder datasets with available demographic metadata. The study involved analysing system performance across various demographic groups, particularly focusing on gender and age-based cohorts. Performance evaluation was based on multiple metrics, including normalised costs and cross-entropy. We employed calibration techniques trained separately on predefined demographic groups to address group-dependent miscalibration. Analysis revealed significant performance disparities across groups despite strong global metrics. The system showed systematic biases, misclassifying healthy speakers over 55 as having a voice disorder and speakers with disorders aged 14-30 as healthy. Group-specific calibration improved posterior probability quality, reducing overconfidence. For young disordered speakers, low severity scores were identified as contributing to poor system performance. For older speakers, age-related voice characteristics and potential limitations in the pretrained Hubert model used as feature extractor likely affected results. The study demonstrates that global performance metrics are insufficient for evaluating AVDD system performance. Group-specific analysis may unmask problems in system performance which are hidden within global metrics. Further, group-dependent calibration strategies help mitigate biases, resulting in a more reliable indication of system confidence. These findings emphasize the need for demographic-specific evaluation and calibration in voice disorder detection systems, while providing a methodological framework applicable to broader biomedical classification tasks where demographic metadata is available.

eess.AS↗

arXiv2017

Measuring $|V_{td}|$ at LHC

We propose a direct measurement of the CKM element $V_{td}$ at the LHC. Taking profit of the imbalance between $d$ and $\bar d$ quark content in the proton, we show that a non-zero $V_{td}$ induces a charge asymmetry in the $tW$ associated production. The main backgrounds to this process, $t\bar t$ production, and $tW$ associated production mediated by $V_{tb}$, give charge symmetric contributions at leading order in QCD. Therefore, using specific kinematic features of the signal, we construct a charge asymmetry in the di-lepton final state which, due also to a reduction of systematic uncertainties in the asymmetry, is potentially sensitive to $V_{td}$ suppressed effects. In particular, using signal and background simulations up to detector level, we show that this new observable could improve the current direct upper bound on $|V_{td}|$ already with existing LHC data. We also project that $|V_{td}|$ values down to $\sim 10$ times the Standard Model prediction could be probed in the high luminosity phase of the LHC.

hep-ph↗

arXiv2017

ttbb as a probe of New Physics at the LHC

We study the $t\bar t b \bar b$ final state at the LHC as a probe of New Physics that couples mainly to third generation of quarks. We analyze New Physics simplified models with resonances of spin 0, 1 and 2. The sensitivity of the final states $t\bar t b\bar b$, $t\bar t t \bar t$, $b \bar b b \bar b$ and $t \bar t$ on each of these models is used to identify an important region in parameter space that is still not excluded and where $t\bar t b \bar b$ is the most sensitive final state. We indicate possible experimental hints and discuss potential issues of observables that rely mainly in Montecarlo predictions. A new observable is proposed that, at the price of requiring more statistics, reduces the impact of Montecarlo predictions. We use preliminary 13 TeV results to give a raw estimate on the discovery reach and propose simple improvements on the observables.

hep-ph↗

arXiv2016

About the isocurvature tension between axion and high scale inflationary models

The present work suggests that the isocurvature tension between axion and high energy inflationary scenarios may be avoided by considering a double field inflationary model involving the hidden Peccei-Quinn Higgs and the Standard Model one. Some terms in the lagrangian we propose explicitly violate the Peccei-Quinn symmetry but, at the present era, their effect is completely negligible. The resulting mechanism allows a large value for the axion constant, of the order $f_a\sim M_p$, thus the axion isocurvature fluctuations are suppressed even when the scale of inflation $H_{inf}$ is very high, of the order of $H_{inf}\sim M_{gut}$. This numerical value is typical in Higgs inflationary models. An analysis about topological defect formation in this scenario is also performed, and it is suggested that, under certain assumptions, their effect is not catastrophic from the cosmological point of view.

hep-ph↗