arXiv subjects

Arnab Das

Publications and source records attributed to Arnab Das.

At least 19 recordsLinked to original sources

arXiv2026

Evidence Subspace Projection: Measuring How Much Evidence Explains Deepfake Detection in Self-Supervised Speech Models

Self-supervised learning (SSL) models are widely used as feature extractors for state-of-the-art audio deepfake detection, but it remains unclear how to directly and quantitatively connect what SSL models capture to detection decisions. To address this gap, we propose Evidence Subspace Projection, a method that represents both evidence factors (e.g., attack category, codec, gender, transmission) and authenticity labels in a shared space constructed from SSL models' neuron activation patterns. By projecting the decision vector onto each evidence subspace, we obtain a scalar ratio that quantifies the explanatory power of each evidence type. We evaluate SSL models in raw, fine-tuned, and post-trained settings on multiple datasets. The results confirm findings from established studies, validating the proposed method, and reveal new insights into model behavior.

cs.SD↗

arXiv2026

DeepFense: A Unified, Modular, and Extensible Framework for Robust Deepfake Audio Detection

Speech deepfake detection is a well-established research field with different models, datasets, and training strategies. However, the lack of standardized implementations and evaluation protocols limits reproducibility, benchmarking, and comparison across studies. In this work, we present DeepFense, a comprehensive, open-source PyTorch toolkit integrating the latest architectures, loss functions, and augmentation pipelines, alongside over 100 recipes. Using DeepFense, we conducted a large-scale evaluation of more than 400 models. Our findings reveal that while carefully curated training data improves cross-domain generalization, the choice of pre-trained front-end feature extractor dominates overall performance variance. Crucially, we show severe biases in high-performing models regarding audio quality, speaker gender, and language. DeepFense is expected to facilitate real-world deployment with the necessary tools to address equitable training data selection and front-end fine-tuning.

cs.SD↗

arXiv2026

DFKI-Speech System for WildSpoof Challenge: A robust framework for SASV In-the-Wild

This paper presents the DFKI-Speech system developed for the WildSpoof Challenge under the Spoofing aware Automatic Speaker Verification (SASV) track. We propose a robust SASV framework in which a spoofing detector and a speaker verification (SV) network operate in tandem. The spoofing detector employs a self-supervised speech embedding extractor as the frontend, combined with a state-of-the-art graph neural network backend. In addition, a top-3 layer based mixture-of-experts (MoE) is used to fuse high-level and low-level features for effective spoofed utterance detection. For speaker verification, we adapt a low-complexity convolutional neural network that fuses 2D and 1D features at multiple scales, trained with the SphereFace loss. Additionally, contrastive circle loss is applied to adaptively weight positive and negative pairs within each training batch, enabling the network to better distinguish between hard and easy sample pairs. Finally, fixed imposter cohort based AS Norm score normalization and model ensembling are used to further enhance the discriminative capability of the speaker verification system.

cs.SD↗

arXiv2026

Charge redistribution at metal-ZrO2 interfaces: A combined DFT and continuum electrostatic study

Nanoscale metallic inclusions (NMIs) are commonly observed within oxide scales formed during high-temperature oxidation, revealing the existence of chemical and electronic heterogeneity beyond conventional corrosion theories that assume homogeneous, fully oxidized films. Using tetragonal zirconia (tZrO2) facing a series of face-centered cubic (fcc) metals as the model system, this work investigates the short-range and long-range charge redistributions across metal-oxide interfaces by coupling density functional theory (DFT) calculations with continuum modeling. We show that metal-oxide contact induces a short-range charge redistribution confined to a few atomic layers and a long-range redistribution of space charge that can extend over macroscopic distances within weakly doped oxides. DFT calculations show that the short-range redistribution is dominated by metal induced gap states (MIGS) in tZrO2 facing noble metals like Au and Ag, and by chemical bonding in tZrO2 facing active metals like Al. DFT-informed continuum theoretical analysis shows that the range of space-charge redistribution is governed by the doping level of tZrO2, and that the Schottky barrier height (SBH) exhibits a stronger dependence on the metal work function than the doping level. Both the short-range and long-range charge redistributions can alter the transport of charge carriers via their associated electric fields, extending several nm to hundreds of nm from the interface, depending on the doping concentrations, suggesting possible heterogeneous oxide growth caused by NMIs.

cond-mat.mtrl-sci↗

arXiv2026

Content Leakage in LibriSpeech and Its Impact on the Privacy Evaluation of Speaker Anonymization

Speaker anonymization aims to conceal a speaker's identity, without considering the linguistic content. In this study, we reveal a weakness of Librispeech, the dataset that is commonly used to evaluate anonymizers: the books read by the Librispeech speakers are so distinct, that speakers can be identified by their vocabularies. Even perfect anonymizers cannot prevent this identity leakage. The EdAcc dataset is better in this regard: only a few speakers can be identified through their vocabularies, encouraging the attacker to look elsewhere for the identities of the anonymized speakers. EdAcc also comprises spontaneous speech and more diverse speakers, complementing Librispeech and giving more insights into how anonymizers work.

eess.AS↗

arXiv2025

Chiral orbital current driven topological Hall effect in Mn3Si2Te6

Chiral orbital current (COC) plays a crucial role in governing the magnetization and transport behaviour in the layered ferrimagnetic nodal-line semiconductor Mn3Si2Te6. Here, we observe that the topological Hall effect (THE), typically attributed to Berry curvature from chiral spin textures, originates from COC, which produces an emergent magnetic field for conduction electrons due to its real-space orbital textures. We find that the THE signal strengthens as we move down from bulk to nanoflakes, but tends to disappear with increasing current, along with the disappearance of the COC state. We also demonstrate a strong correlation between the colossal magnetoresistance (CMR) and the observed THE, suggesting that large Berry curvature and topological transport can arise purely from orbital degrees of freedom, providing a new platform for engineering dissipationless transport in 2D magnets.

cond-mat.other↗

arXiv2025

A Parameter-Efficient Multi-Scale Convolutional Adapter for Synthetic Speech Detection

Recent synthetic speech detection models typically adapt a pre-trained SSL model via finetuning, which is computationally demanding. Parameter-Efficient Fine-Tuning (PEFT) offers an alternative. However, existing methods lack the specific inductive biases required to model the multi-scale temporal artifacts characteristic of spoofed audio. This paper introduces the Multi-Scale Convolutional Adapter (MultiConvAdapter), a parameter-efficient architecture designed to address this limitation. MultiConvAdapter integrates parallel convolutional modules within the SSL encoder, facilitating the simultaneous learning of discriminative features across multiple temporal resolutions, capturing both short-term artifacts and long-term distortions. With only $3.17$M trainable parameters ($1\%$ of the SSL backbone), MultiConvAdapter substantially reduces the computational burden of adaptation. Evaluations on five public datasets, demonstrate that MultiConvAdapter achieves superior performance compared to full fine-tuning and established PEFT methods.

cs.SD↗

arXiv2025

Generalizable Audio Spoofing Detection using Non-Semantic Representations

Rapid advancements in generative modeling have made synthetic audio generation easy, making speech-based services vulnerable to spoofing attacks. Consequently, there is a dire need for robust countermeasures more than ever. Existing solutions for deepfake detection are often criticized for lacking generalizability and fail drastically when applied to real-world data. This study proposes a novel method for generalizable spoofing detection leveraging non-semantic universal audio representations. Extensive experiments have been performed to find suitable non-semantic features using TRILL and TRILLsson models. The results indicate that the proposed method achieves comparable performance on the in-domain test set while significantly outperforming state-of-the-art approaches on out-of-domain test sets. Notably, it demonstrates superior generalization on public-domain data, surpassing methods based on hand-crafted features, semantic embeddings, and end-to-end architectures.

cs.SD↗

arXiv2025

Improving the Speaker Anonymization Evaluation's Robustness to Target Speakers with Adversarial Learning

The current privacy evaluation for speaker anonymization often overestimates privacy when a same-gender target selection algorithm (TSA) is used, although this TSA leaks the speaker's gender and should hence be more vulnerable. We hypothesize that this occurs because the evaluation does not account for the fact that anonymized speech contains information from both the source and target speakers. To address this, we propose to add a target classifier that measures the influence of target speaker information in the evaluation, which can also be removed with adversarial learning. Experiments demonstrate that this approach is effective for multiple anonymizers, particularly when using a same-gender TSA, leading to a more reliable assessment.

eess.AS↗

arXiv2025

Two Views, One Truth: Spectral and Self-Supervised Features Fusion for Robust Speech Deepfake Detection

Recent advances in synthetic speech have made audio deepfakes increasingly realistic, posing significant security risks. Existing detection methods that rely on a single modality, either raw waveform embeddings or spectral based features, are vulnerable to non spoof disturbances and often overfit to known forgery algorithms, resulting in poor generalization to unseen attacks. To address these shortcomings, we investigate hybrid fusion frameworks that integrate self supervised learning (SSL) based representations with handcrafted spectral descriptors (MFCC , LFCC, CQCC). By aligning and combining complementary information across modalities, these fusion approaches capture subtle artifacts that single feature approaches typically overlook. We explore several fusion strategies, including simple concatenation, cross attention, mutual cross attention, and a learnable gating mechanism, to optimally blend SSL features with fine grained spectral cues. We evaluate our approach on four challenging public benchmarks and report generalization performance. All fusion variants consistently outperform an SSL only baseline, with the cross attention strategy achieving the best generalization with a 38% relative reduction in equal error rate (EER). These results confirm that joint modeling of waveform and spectral views produces robust, domain agnostic representations for audio deepfake detection.

cs.SD↗

arXiv2025

Periodic Drive Induced Half-Metallic Phase in Insulators and Correlated Metals

Non-equilibrium control of electronic properties in condensed matter systems can result in novel phenomena. In this work, we provide a novel non-equilibrium route to realize half-metallic phases. We explore the periodically driven Hubbard model on a bipartite lattice and demonstrate that a periodic drive can transform a weakly interacting metal into a ferrimagnetic half-metal. We consider a Fermi-Hubbard model with only nearest-neighbour hopping and stabilize the elusive phase simply by driving the site potentials periodically. The drive induces staggered second and third-neighbor hopping and a staggered potential between two sublattices in the Floquet Hamiltonian, whose ground state is explored in this work. Close to the dynamical freezing point, due to the suppression of nearest neighbor hopping in the driven system, an effective enhancement of various terms in the Floquet Hamiltonian, including the e-e interactions, occurs. This helps in stabilizing a broad ferrimagnetic half-metallic phase for a wide range of system parameters. The half-metallic phase achieved in the presence of high drive frequency should be stable for exponentially large time scales in drive frequency and could be perpetually stable beyond a strong enough drive amplitude owing to dynamical freezing. It can hence have potential applications in stable spintronics and other upcoming quantum technologies.

cond-mat.str-el↗

arXiv2025

Machine Learning-Based Anomaly Detection of Correlated Sensor Data: An Integrated Principal Component Analysis-Autoencoder Approach

The growing adoption of IoT systems in industries like transportation, banking, healthcare, and smart energy has increased reliance on sensor networks. However, anomalies in sensor readings can undermine system reliability, making real-time anomaly detection essential. While a large body of research addresses anomaly detection in IoT networks, few studies focus on correlated sensor data streams, such as temperature and pressure within a shared space, especially in resource-constrained environments. To address this, we propose a novel hybrid machine learning approach combining Principal Component Analysis (PCA) and Autoencoders. In this method, PCA continuously monitors sensor data and triggers the Autoencoder when significant variations are detected. This hybrid approach, validated with real-world and simulated data, shows faster response times and fewer false positives. The F1 score of the hybrid method is comparable to Autoencoder, with much faster response time which is driven by PCA.

eess.SP↗

arXiv2025

Private kNN-VC: Interpretable Anonymization of Converted Speech

Speaker anonymization seeks to conceal a speaker's identity while preserving the utility of their speech. The achieved privacy is commonly evaluated with a speaker recognition model trained on anonymized speech. Although this represents a strong attack, it is unclear which aspects of speech are exploited to identify the speakers. Our research sets out to unveil these aspects. It starts with kNN-VC, a powerful voice conversion model that performs poorly as an anonymization system, presumably because of prosody leakage. To test this hypothesis, we extend kNN-VC with two interpretable components that anonymize the duration and variation of phones. These components increase privacy significantly, proving that the studied prosodic factors encode speaker identity and are exploited by the privacy attack. Additionally, we show that changes in the target selection algorithm considerably influence the outcome of the privacy attack.

eess.AS↗

arXiv2025

Satire: Computing Rigorous Bounds for Floating-Point Rounding Error in Mixed-Precision Loop-Free Programs

Techniques that rigorously bound the overall rounding error exhibited by a numerical program are of significant interest for communities developing numerical software. However, there are few available tools today that can be used to rigorously bound errors in programs that employ conditional statements (a basic need) as well as mixed-precision arithmetic (a direction of significant future interest) employing global optimization in error analysis. In this paper, we present a new tool that fills this void while also employing an abstraction-guided optimization approach to allow designers to trade error-bound tightness for gains in analysis time -- useful when searching for design alternatives. We first present the basic rigorous analysis framework of Satire and then show how to extend it to incorporate abstractions, conditionals, and mixed-precision arithmetic. We begin by describing Satire's design and its performance on a collection of benchmark examples. We then describe these aspects of Satire: (1) how the error-bound and tool execution time vary with the abstraction level; (2) the additional machinery to handle conditional expression branches, including defining the concepts of instability jumps and instability window widths and measuring these quantities; and (3) how the error changes when a mix of precision values are used. To showcase how \satire can add value during design, we start with a Conjugate Gradient solver and demonstrate how its step size and search direction are affected by different precision settings. Satire is freely available for evaluation, and can be used during the design of numerical routines to effect design tradeoffs guided by rigorous empirical error guarantees.

cs.PL↗

arXiv2025

Tuning the chiral orbital currents in a colossal magnetoresistive nodal line ferrimagnet

The ferrimagnetic nodal-line semiconductor Mn$_3$Si$_2$Te$_6$ exhibits colossal magnetoresistance (CMR) owing to the chiral orbital currents (COC). The COC is developed due to spin-orbit interaction (SOI) attributed to the tellurium (Te) atoms. Here, we observe that on chemical substitution of the Te atoms with selenium (Se), the COC, which runs along the Te-Te edges of the MnTe$_6$ octahedra, becomes weaker and thus affects the angular magnetoresistance (MR) of Mn$_3$Si$_2$Te$_6$. We find that the application of magnetic field along the easy axis leads to a considerable drop in resistance in substituted crystals, which otherwise exhibits weak MR. On the other hand, the CMR effect along the partially polarized magnetization direction is found to be only marginally affected due to the substitution and persists even for a significantly high concentration of Se.

cond-mat.mtrl-sci↗

arXiv2024

Anonymising Elderly and Pathological Speech: Voice Conversion Using DDSP and Query-by-Example

Speech anonymisation aims to protect speaker identity by changing personal identifiers in speech while retaining linguistic content. Current methods fail to retain prosody and unique speech patterns found in elderly and pathological speech domains, which is essential for remote health monitoring. To address this gap, we propose a voice conversion-based method (DDSP-QbE) using differentiable digital signal processing and query-by-example. The proposed method, trained with novel losses, aids in disentangling linguistic, prosodic, and domain representations, enabling the model to adapt to uncommon speech patterns. Objective and subjective evaluations show that DDSP-QbE significantly outperforms the voice conversion state-of-the-art concerning intelligibility, prosody, and domain preservation across diverse datasets, pathologies, and speakers while maintaining quality and speaker anonymity. Experts validate domain preservation by analysing twelve clinically pertinent domain attributes.

cs.AI↗

arXiv2024

Dynamical freezing in the thermodynamic limit: the strongly driven ensemble

The ergodicity postulate, a foundational pillar of Gibbsian statistical mechanics predicts that a periodically driven (Floquet) system in the absence of any conservation law heats to a featureless `infinite temperature' state. Here, we find--for a clean and interacting generic spin chain subject to a {\it strong} driving field--that this can be prevented by the emergence of {\it approximate but stable} conservation-laws not present in the undriven system. We identify their origin: they do not necessarily owe their stability to familiar protections by symmetry, topology, disorder, or even high energy costs. We show numerically, {\it in the thermodynamic limit,} that when required by these emergent conservation-laws, the entanglement-entropy density of an infinite subsystem remains zero over our entire simulation time of several decades in natural units. We further provide a recipe for designing such conservation laws with high accuracy. Finally, we present an ensemble description, which we call the strongly driven ensemble incorporating these constraints. This provides a way to control many-body chaos through stable Floquet-engineering. Strong signatures of these conservation-laws should be experimentally accessible since they manifest in all length and time scales. Variants of the spin model we have used, have already been realized using Rydberg-dressed atoms.

cond-mat.stat-mech↗

arXiv2024

Stretched-Exponential Melting of a Dynamically Frozen State Under Imprinted Phase Noise in the Ising Chain in a Transverse Field

Dynamical freezing is a phenomenon where a set of local observables emerges as approximate but stable conserved quantities (freezes) under a strong periodic drive in a closed quantum system. The expectation values of these emergent conserved quantities exhibit small fluctuations around their respective initial values. These fluctuations do not grow with time, and their magnitude can be tuned down sharply by tuning the drive parameters. In this work, we probe the resilience of dynamical freezing to random perturbations added to the relative phases between the interfering states (elements of a natural basis) in the time-evolving wave function after each drive cycle. We study this in an integrable Ising chain in a time-periodic transverse field. Our key finding is, that the imprinted phase noise melts the dynamically frozen state, but the decay is "slow": a stretched-exponential decay rather than an exponential one. Stretched-exponential decays (also known as Kohlrausch relaxation) are usually expected in complex systems with time-scale hierarchies due to strong disorders or other inhomogeneities resulting in jamming, glassiness, or localization.

cond-mat.stat-mech↗