arXiv subjects

Arnab Das

Publications and source records attributed to Arnab Das.

At least 37 records · Page 2Linked to original sources

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↗

arXiv2023

PI-RADS v2 Compliant Automated Segmentation of Prostate Zones Using co-training Motivated Multi-task Dual-Path CNN

The detailed images produced by Magnetic Resonance Imaging (MRI) provide life-critical information for the diagnosis and treatment of prostate cancer. To provide standardized acquisition, interpretation and usage of the complex MRI images, the PI-RADS v2 guideline was proposed. An automated segmentation following the guideline facilitates consistent and precise lesion detection, staging and treatment. The guideline recommends a division of the prostate into four zones, PZ (peripheral zone), TZ (transition zone), DPU (distal prostatic urethra) and AFS (anterior fibromuscular stroma). Not every zone shares a boundary with the others and is present in every slice. Further, the representations captured by a single model might not suffice for all zones. This motivated us to design a dual-branch convolutional neural network (CNN), where each branch captures the representations of the connected zones separately. Further, the representations from different branches act complementary to each other at the second stage of training, where they are fine-tuned through an unsupervised loss. The loss penalises the difference in predictions from the two branches for the same class. We also incorporate multi-task learning in our framework to further improve the segmentation accuracy. The proposed approach improves the segmentation accuracy of the baseline (mean absolute symmetric distance) by 7.56%, 11.00%, 58.43% and 19.67% for PZ, TZ, DPU and AFS zones respectively.

eess.IV↗

arXiv2023

StarGAN-VC++: Towards Emotion Preserving Voice Conversion Using Deep Embeddings

Voice conversion (VC) transforms an utterance to sound like another person without changing the linguistic content. A recently proposed generative adversarial network-based VC method, StarGANv2-VC is very successful in generating natural-sounding conversions. However, the method fails to preserve the emotion of the source speaker in the converted samples. Emotion preservation is necessary for natural human-computer interaction. In this paper, we show that StarGANv2-VC fails to disentangle the speaker and emotion representations, pertinent to preserve emotion. Specifically, there is an emotion leakage from the reference audio used to capture the speaker embeddings while training. To counter the problem, we propose novel emotion-aware losses and an unsupervised method which exploits emotion supervision through latent emotion representations. The objective and subjective evaluations prove the efficacy of the proposed strategy over diverse datasets, emotions, gender, etc.

eess.AS↗

arXiv2023

Emo-StarGAN: A Semi-Supervised Any-to-Many Non-Parallel Emotion-Preserving Voice Conversion

Speech anonymisation prevents misuse of spoken data by removing any personal identifier while preserving at least linguistic content. However, emotion preservation is crucial for natural human-computer interaction. The well-known voice conversion technique StarGANv2-VC achieves anonymisation but fails to preserve emotion. This work presents an any-to-many semi-supervised StarGANv2-VC variant trained on partially emotion-labelled non-parallel data. We propose emotion-aware losses computed on the emotion embeddings and acoustic features correlated to emotion. Additionally, we use an emotion classifier to provide direct emotion supervision. Objective and subjective evaluations show that the proposed approach significantly improves emotion preservation over the vanilla StarGANv2-VC. This considerable improvement is seen over diverse datasets, emotions, target speakers, and inter-group conversions without compromising intelligibility and anonymisation.

eess.AS↗

arXiv2022

TorchEsegeta: Framework for Interpretability and Explainability of Image-based Deep Learning Models

Clinicians are often very sceptical about applying automatic image processing approaches, especially deep learning based methods, in practice. One main reason for this is the black-box nature of these approaches and the inherent problem of missing insights of the automatically derived decisions. In order to increase trust in these methods, this paper presents approaches that help to interpret and explain the results of deep learning algorithms by depicting the anatomical areas which influence the decision of the algorithm most. Moreover, this research presents a unified framework, TorchEsegeta, for applying various interpretability and explainability techniques for deep learning models and generate visual interpretations and explanations for clinicians to corroborate their clinical findings. In addition, this will aid in gaining confidence in such methods. The framework builds on existing interpretability and explainability techniques that are currently focusing on classification models, extending them to segmentation tasks. In addition, these methods have been adapted to 3D models for volumetric analysis. The proposed framework provides methods to quantitatively compare visual explanations using infidelity and sensitivity metrics. This framework can be used by data scientists to perform post-hoc interpretations and explanations of their models, develop more explainable tools and present the findings to clinicians to increase their faith in such models. The proposed framework was evaluated based on a use case scenario of vessel segmentation models trained on Time-of-fight (TOF) Magnetic Resonance Angiogram (MRA) images of the human brain. Quantitative and qualitative results of a comparative study of different models and interpretability methods are presented. Furthermore, this paper provides an extensive overview of several existing interpretability and explainability methods.

cs.CV↗

arXiv2021

Signatures of quantum phase transitions after quenches in quantum chaotic one-dimensional systems

Quantum phase transitions are central to our understanding of why matter at very low temperatures can exhibit starkly different properties upon small changes of microscopic parameters. Accurately locating those transitions is challenging experimentally and theoretically. Here we show that the antithetic strategy of forcing systems out of equilibrium via sudden quenches provides a route to locate quantum phase transitions. Specifically, we show that such transitions imprint distinctive features in the intermediate-time dynamics, and results after equilibration, of local observables in quantum-chaotic spin chains. Furthermore, we show that the effective temperature in the expected thermal-like states after equilibration can exhibit minima in the vicinity of the quantum critical points. We discuss how to test our results in experiments with Rydberg atoms, and explore nonequilibrium signatures of quantum critical points in models with topological transitions.

cond-mat.stat-mech↗

arXiv2021

Dynamics and correlations at a quantum phase transition beyond Kibble-Zurek

Kibble-Zurek theory (KZ) stands out as the most robust theory of defect generation in the dynamics of phase transitions. KZ utilizes the structure of equilibrium states away from the transition point to estimate the excitations due to the transition using adiabatic and impulse approximations. Here we show, the actual nonequilibrium dynamics lead to a qualitatively different scenario from KZ, as far correlations between the defects (rather than their densities) are concerned. For a quantum Ising chain, we show, this gives rise to a Gaussian spatial decay in the domain wall (kinks) correlations, while KZ would predict an exponential fall. We propose a simple but general framework on top of KZ, based on the "quantum coarsening" dynamics of local correlators in the supposed impulse regime. We outline how our picture extends to generic interacting situations.

cond-mat.str-el↗

arXiv2021

Statistical Mechanics of Floquet Quantum Matter: Exact and Emergent Conservation Laws

Equilibrium statistical mechanics rests on the assumption of ergodic dynamics of a system modulo the conservation laws of local observables: extremization of entropy immediately gives Gibbs' ensemble (GE) for energy conserving systems and a generalized version of it (GGE) when the number of local conserved quantities (LCQ) is more than one. Through the last decade, statistical mechanics has been extended to describe the late-time behaviour of periodically driven (Floquet) quantum matter starting from a generic state. The structure built on the fundamental assumptions of ergodicity and identification of the relevant "conservation laws" in this inherently non-equilibrium setting. More recently, it has been shown that the statistical mechanics has a much richer structure due to the existence of {\it emergent} conservation laws: these are approximate but stable conservation laws arising {\it due to the drive}, and are not present in the undriven system. Extensive numerical and analytical results support perpetual stability of these emergent (though approximate) conservation laws, probably even in the thermodynamic limit. This banks on the recent finding of a sharp ergodicity threshold for Floquet thermalization in clean, interacting non-integrable Floquet systems. This opens up a new possibility of stable Floquet engineering in such systems. This review intends to give a theoretical overview of these developments. We conclude by briefly surveying the experimental scenario.

cond-mat.stat-mech↗

arXiv2021

Dynamical Freezing and Scar Points in Strongly Driven Floquet Matter: Resonance vs Emergent Conservation Laws

We consider a clean quantum system subject to strong periodic driving. The existence of a dominant energy scale, $h_D^x$, can generate considerable structure in an effective description of a system which, in the absence of the drive, is non-integrable, interacting, and does not host localization. In particular, we uncover points of freezing in the space of drive parameters (frequency and amplitude). At those points, the dynamics is severely constrained due to the emergence of an almost exact local conserved quantity, which scars the {\it entire} Floquet spectrum by preventing the system from heating up ergodically, starting from any generic state, even though it delocalizes over an appropriate subspace. At large drive frequencies, where a naïve Magnus expansion would predict a vanishing effective (average) drive, we devise instead a strong-drive Magnus expansion in a moving frame. There, the emergent conservation law is reflected in the appearance of an `integrability' of an effective Hamiltonian. These results hold for a wide variety of Hamiltonians, including the Ising model in a transverse field in {\it any dimension} and for {\it any form of Ising interactions}. The phenomenon is also shown to be robust in the presence of {\it two-body Heisenberg interactions with any arbitrary choice of couplings}. Further, we construct a real-time perturbation theory which captures resonance phenomena where the conservation breaks down, giving way to unbounded heating. This opens a window on the low-frequency regime where the Magnus expansion fails.

cond-mat.other↗

arXiv2020

Disorder-induced Enhancement of Entanglement Growth in One Dimension: Information Leakage at the scale of localization length

When a group of compactly packed free fermions is allowed to spread over an empty one-dimensional lattice, the spreading particles can create entanglement between different parts of the lattice. We show, though breaking of translational invariance (TI) of the lattice by disorder slows down the spreading of local observables, the entanglement entropy of a subsystem can nonetheless receive a remarkable enhancement as long as the subsystem lies within the single-particle localization length. We show, the main mechanism behind this enhancement is the re-entrant exchange of particles between the subparts due to transport of mutual information due to back scattering. We discuss the length and time scales relevant to the phenomenon. We study the phenomenon for breaking of TI by both quasi-periodic and random potentials. We further explore the effect of randomness only in the initial state. This also exhibits a similar enhancement effect even in a TI lattice. We also touch upon the special case of periodic potential, where qualitatively similar phenomenology emerges, though the coherence in the back scattering in this case leads to effects not captured by our simple yet generic picture.

cond-mat.str-el↗

arXiv2020

FPDetect: Efficient Reasoning About Stencil Programs Using Selective Direct Evaluation

We present FPDetect, a low overhead approach for detecting logical errors and soft errors affecting stencil computations without generating false positives. We develop an offline analysis that tightly estimates the number of floating-point bits preserved across stencil applications. This estimate rigorously bounds the values expected in the data space of the computation. Violations of this bound can be attributed with certainty to errors. FPDetect helps synthesize error detectors customized for user-specified levels of accuracy and coverage. FPDetect also enables overhead reduction techniques based on deploying these detectors coarsely in space and time. Experimental evaluations demonstrate the practicality of our approach.

cs.DC↗

arXiv2020

An Abstraction-guided Approach to Scalable and Rigorous Floating-Point Error Analysis

Automated techniques for rigorous floating-point round-off error analysis are important in areas including formal verification of correctness and precision tuning. Existing tools and techniques, while providing tight bounds, fail to analyze expressions with more than a few hundred operators, thus unable to cover important practical problems. In this work, we present Satire, a new tool that sheds light on how scalability and bound-tightness can be attained through a combination of incremental analysis, abstraction, and judicious use of concrete and symbolic evaluation. Satire has handled problems exceeding 200K operators. We present Satire's underlying error analysis approach, information-theoretic abstraction heuristics, and a wide range of case studies, with evaluation covering FFT, Lorenz system of equations, and various PDE stencil types. Our results demonstrate the tightness of Satire's bounds, its acceptable runtime, and valuable insights provided.

cs.PL↗

arXiv2018

Multifractality without fine-tuning in a Floquet quasiperiodic chain

Periodically driven, or Floquet, disordered quantum systems have generated many unexpected discoveries of late, such as the anomalous Floquet Anderson insulator and the discrete time crystal. Here, we report the emergence of an entire band of multifractal wavefunctions in a periodically driven chain of non-interacting particles subject to spatially quasiperiodic disorder. Remarkably, this multifractality is robust in that it does not require any fine-tuning of the model parameters, which sets it apart from the known multifractality of $critical$ wavefunctions. The multifractality arises as the periodic drive hybridises the localised and delocalised sectors of the undriven spectrum. We account for this phenomenon in a simple random matrix based theory. Finally, we discuss dynamical signatures of the multifractal states, which should betray their presence in cold atom experiments. Such a simple yet robust realisation of multifractality could advance this so far elusive phenomenon towards applications, such as the proposed disorder-induced enhancement of a superfluid transition.

cond-mat.dis-nn↗

arXiv2018

Onset of Floquet Thermalisation

In presence of interactions, a closed, homogeneous (disorder-free) many-body system is believed to generically heat up to an `infinite temperature' ensemble when subjected to a periodic drive: in the spirit of the ergodicity hypothesis underpinning statistical mechanics, this happens as no energy or other conservation law prevents this. Here we present an interacting Ising chain driven by a field of time-dependent strength, where such heating onsets only below a threshold value of the drive amplitude, above which the system exhibits non-ergodic behaviour. The onset appears at {\it strong, but not fast} driving. This in particular puts it beyond the scope of high-frequency expansions. The onset location shifts, but it is robustly present, across wide variations of the model Hamiltonian such as driving frequency and protocol, as well as the initial state. The portion of nonergodic states in the Floquet spectrum, while thermodynamically subdominant, has a finite entropy. We find that the magnetisation as an {\it emergent} conserved quantity underpinning the freezing; indeed the freezing effect is readily observed, as initially magnetised states remain partially frozen {\it up to infinite time}. This result, which bears a family resemblance to the Kolmogorov-Arnold-Moser theorem for classical dynamical systems, could be a valuable ingredient for extending Floquet engineering to the interacting realm.

cond-mat.stat-mech↗

arXiv2017

Dynamical Many-body Localization and Delocalization in Periodically Driven Closed Quantum Systems

Quantum interference lies at the heart of several surprising equilibrium and non-equilibrium phenomena in many-body Physics. Here we discuss two recently explored non-equilibrium scenarios where external periodic drive applied to closed (i.e., not attached to any external bath) quantum many-body systems have apparently opposite effects in respective cases. In one case it freezes/localizes a disorder free system dynamically, while in the other it delocalizes a disordered many-body localized system, and quantum interference is responsible for both the effects. We review these in the perspective of more general questions of ergodicity, energy absorption, asymptotic behavior, and finally the essential role of quantum mechanics in understanding these issues in periodically driven closed many-body systems. In this article we intend to deliver a non-technical account of some recent developments in this field in a manner accessible to a broad readership.

cond-mat.other↗

arXiv2017

Locating topological phase transitions using nonequilibrium signatures in local bulk observables

Topological quantum phases cannot be characterized by local order parameters in the bulk. In this work however, we show that signatures of a topological quantum critical point do remain in local observables in the bulk, and manifest themselves as non-analyticities in their expectation values taken over a family of non-equilibrium states generated using a quantum quench protocol. The signature can be used for precisely locating the critical points in parameter space. A large class of initial states can be chosen for the quench (including finite temperature states), the sufficient condition being existence of a finite occupation-gradient with respect to energy for the single-particle critical mode. We demonstrate these results in tractable models of non-interacting fermions exhibiting topological phase transitions in one and two spatial dimensions. We also show that the non-analyticities can be absent if the gap-closing is non-topological, i.e., when it corresponds to no phase transition.

cond-mat.stat-mech↗

arXiv2016

Eigenstate Gibbs Ensemble in Integrable Quantum Systems

The Eigenstate Thermalization Hypothesis implies that for a thermodynamically large system in one of its eigenstates, the reduced density matrix describing any finite subsystem is determined solely by a set of {\it relevant} conserved quantities. In a generic system, only the energy plays that role and hence eigenstates appear locally thermal. Integrable systems, on the other hand, possess an extensive number of such conserved quantities and hence the reduced density matrix requires specification of an infinite number of parameters (Generalized Gibbs Ensemble). However, here we show by unbiased statistical sampling of the individual eigenstates with a given finite energy density, that the local description of an overwhelming majority of these states of even such an integrable system is actually Gibbs-like, i.e. requires only the energy density of the eigenstate. Rare eigenstates that cannot be represented by the Gibbs ensemble can also be sampled efficiently by our method and their local properties are then shown to be described by appropriately {\it truncated} Generalized Gibbs Ensembles. We further show that the presence of these rare eigenstates differentiates the model from the generic (non-integrable) case and leads to the system being described by a Generalized Gibbs Ensemble at long time under a unitary dynamics following a sudden quench, even when the initial state is a Gibbs-like eigenstate of the pre-quench Hamiltonian.

cond-mat.stat-mech↗