arXiv subjects

Haim Permuter

Publications and source records attributed to Haim Permuter.

At least 19 recordsLinked to original sources

arXiv2026

Multi Codec Discrete Diffusion Model for Text Guided Speech Inpainting and Editing

Speech recordings often contain missing, corrupted, or incorrect regions that must be reconstructed or modified without re-synthesizing the entire utterance. Speech inpainting restores missing segments, whereas speech editing replaces spoken content according to an edited transcript. Both tasks require the generated speech to express the intended words while remaining consistent with the surrounding speaker identity, prosody, timing, and recording conditions. Discrete diffusion is particularly well suited to these tasks because it can iteratively refine masked tokens while jointly conditioning on both left and right acoustic context. We introduce SIEDD, a discrete diffusion framework for text-guided speech inpainting and editing over hierarchical codec tokens. Its core architecture, HiCoDD, follows the RVQ generation order by representing previously generated codebooks as clean, committed acoustic context and applying diffusion only to the current refinement codebook. This separation enables leakage-free joint training while matching sequential coarse-to-fine inference. The model further combines phoneme-level conditioning, span-localized classifier-free guidance, and duration prediction to support both fixed-duration inpainting and variable-duration text edits. On the RealEdit benchmark, SIEDD achieves the best overall speech-editing performance among the evaluated methods. It also outperforms the evaluated autoregressive baselines across all speech-inpainting settings, on both single and multiple gaps. These results demonstrate that explicitly modeling the codec hierarchy substantially improves context-preserving speech reconstruction and editing. See our full code at https://github.com/iftachShoham/SIEDD.

cs.SD↗

arXiv2026

Split and Conquer Partial Deepfake Speech

Partial deepfake speech detection requires identifying manipulated regions that may occur within short temporal portions of an otherwise bona fide utterance, making the task particularly challenging for conventional utterance-level classifiers. We propose a split-and-conquer framework that decomposes the problem into two stages: boundary detection and segment-level classification. A dedicated boundary detector first identifies temporal transition points, allowing the audio signal to be divided into segments that are expected to contain acoustically consistent content. Each resulting segment is then evaluated independently to determine whether it corresponds to bona fide or fake speech. This formulation simplifies the learning objective by explicitly separating temporal localization from authenticity assessment, allowing each component to focus on a well-defined task. To further improve robustness, we introduce a reflection-based multi-length training strategy that converts variable-duration segments into several fixed input lengths, producing diverse feature-space representations. Each stage is trained using multiple configurations with different feature extractors and augmentation strategies, and their complementary predictions are fused to obtain improved final models. Experiments on the PartialSpoof benchmark demonstrate state-of-the-art performance across multiple temporal resolutions as well as at the utterance level, with substantial improvements in the accurate detection and localization of spoofed regions. In addition, the proposed method achieves state-of-the-art performance on the Half-Truth dataset, further confirming the robustness and generalization capability of the framework.

cs.SD↗

arXiv2026

Directed Information: Estimation, Optimization and Applications in Communications and Causality

Directed information (DI) is an information measure that attempts to capture directionality in the flow of information from one random process to another. It is closely related to other causal influence measures, such as transfer entropy, Granger causality, and Pearl's causal framework. This monograph provides an overview of DI and its main application in information theory, namely, characterizing the capacity of channels with feedback and memory. We begin by reviewing the definitions of DI, its basic properties, and its relation to Shannon's mutual information. Next, we provide a survey of DI estimation techniques, ranging from classic plug-in estimators to modern neural-network-based estimators. Considering the application of channel capacity estimation, we describe how such estimators numerically optimize DI rate over a class of joint distributions on input and output processes. A significant part of the monograph is devoted to techniques to compute the feedback capacity of finite-state channels (FSCs). The feedback capacity of a strongly connected FSC involves the maximization of the DI rate from the channel input process to the output process. This maximization is performed over the class of causal conditioned probability input distributions. When the FSC is also unifilar, i.e., the next state is given by a time-invariant function of the current state and the new input-output symbol pair, the feedback capacity is the optimal average reward of an appropriately formulated Markov decision process (MDP). This MDP formulation has been exploited to develop several methods to compute exactly, or at least estimate closely, the feedback capacity of a unifilar FSC. This monograph describes these methods, starting from the value iteration algorithm, to Q-graph methods, and reinforcement learning algorithms that can handle large input and output alphabets.

cs.IT↗

arXiv2025

Unmasking Deepfakes: Leveraging Augmentations and Features Variability for Deepfake Speech Detection

Deepfake speech detection presents a growing challenge as generative audio technologies continue to advance. We propose a hybrid training framework that advances detection performance through novel augmentation strategies. First, we introduce a dual-stage masking approach that operates both at the spectrogram level (MaskedSpec) and within the latent feature space (MaskedFeature), providing complementary regularization that improves tolerance to localized distortions and enhances generalization learning. Second, we introduce compression-aware strategy during self-supervised to increase variability in low-resource scenarios while preserving the integrity of learned representations, thereby improving the suitability of pretrained features for deepfake detection. The framework integrates a learnable self-supervised feature extractor with a ResNet classification head in a unified training pipeline, enabling joint adaptation of acoustic representations and discriminative patterns. On the ASVspoof5 Challenge (Track~1), the system achieves state-of-the-art results with an Equal Error Rate (EER) of 4.08% under closed conditions, further reduced to 2.71% through fusion of models with diverse pretrained feature extractors. when trained on ASVspoof2019, our system obtaining leading performance on the ASVspoof2019 evaluation set (0.18% EER) and the ASVspoof2021 DF task (2.92% EER).

cs.SD↗

arXiv2025

Robust Localization of Partially Fake Speech: Metrics and Out-of-Domain Evaluation

Partial audio deepfake localization poses unique challenges and remain underexplored compared to full-utterance spoofing detection. While recent methods report strong in-domain performance, their real-world utility remains unclear. In this analysis, we critically examine the limitations of current evaluation practices, particularly the widespread use of Equal Error Rate (EER), which often obscures generalization and deployment readiness. We propose reframing the localization task as a sequential anomaly detection problem and advocate for the use of threshold-dependent metrics such as accuracy, precision, recall, and F1-score, which better reflect real-world behavior. Specifically, we analyze the performance of the open-source Coarse-to-Fine Proposal Refinement Framework (CFPRF), which achieves a 20-ms EER of 7.61% on the in-domain PartialSpoof evaluation set, but 43.25% and 27.59% on the LlamaPartialSpoof and Half-Truth out-of-domain test sets. Interestingly, our reproduced version of the same model performs worse on in-domain data (9.84%) but better on the out-of-domain sets (41.72% and 14.98%, respectively). This highlights the risks of over-optimizing for in-domain EER, which can lead to models that perform poorly in real-world scenarios. It also suggests that while deep learning models can be effective on in-domain data, they generalize poorly to out-of-domain scenarios, failing to detect novel synthetic samples and misclassifying unfamiliar bona fide audio. Finally, we observe that adding more bona fide or fully synthetic utterances to the training data often degrades performance, whereas adding partially fake utterances improves it.

cs.SD↗

arXiv2025

TREET: TRansfer Entropy Estimation via Transformers

Transfer entropy (TE) is an information theoretic measure that reveals the directional flow of information between processes, providing valuable insights for a wide range of real-world applications. This work proposes Transfer Entropy Estimation via Transformers (TREET), a novel attention-based approach for estimating TE for stationary processes. The proposed approach employs Donsker-Varadhan representation to TE and leverages the attention mechanism for the task of neural estimation. We propose a detailed theoretical and empirical study of the TREET, comparing it to existing methods on a dedicated estimation benchmark. To increase its applicability, we design an estimated TE optimization scheme that is motivated by the functional representation lemma, and use it to estimate the capacity of communication channels with memory, which is a canonical optimization problem in information theory. We further demonstrate how an optimized TREET can be used to estimate underlying densities, providing experimental results. Finally, we apply TREET to feature analysis of patients with Apnea, demonstrating its applicability to real-world physiological data. Our work, applied with state-of-the-art deep learning methods, opens a new door for communication problems which are yet to be solved.

cs.IT↗

arXiv2025

Token-Based Audio Inpainting via Discrete Diffusion

Audio inpainting seeks to restore missing segments in degraded recordings. Previous diffusion-based methods exhibit impaired performance when the missing region is large. We introduce the first approach that applies discrete diffusion over tokenized music representations from a pre-trained audio tokenizer, enabling stable and semantically coherent restoration of long gaps. Our method further incorporates two training approaches: a derivative-based regularization loss that enforces smooth temporal dynamics, and a span-based absorbing transition that provides structured corruption during diffusion. Experiments on the MusicNet and MAESTRO datasets with gaps up to 750 ms show that our approach consistently outperforms strong baselines across range of gap lengths, for gaps of 150 ms and above. This work advances musical audio restoration and introduces new directions for discrete diffusion model training. Visit our project page for examples and code.

cs.SD↗

arXiv2025

Plan for Speed: Dilated Scheduling for Masked Diffusion Language Models

Masked diffusion language models (MDLMs) promise fast, non-autoregressive text generation, yet existing samplers, which pick tokens to unmask based on model confidence, ignore interactions when unmasking multiple positions in parallel and effectively reduce to slow, autoregressive behavior. We propose the Dilated Unmasking Scheduler (DUS), an inference-only, planner-model-free method that partitions sequence positions into non-adjacent dilated groups and unmasks them in parallel so as to minimize an upper bound on joint entropy gain at each denoising step. By explicitly trading off the number of network calls against generation quality, DUS recovers most of the performance lost under traditional parallel unmasking strategies. Across math (GSM8K, MATH500), code (HumanEval, MBPP), general-knowledge (BBH, MMLU-Pro), and instruction following (IFEval) benchmarks, DUS outperforms confidence-based planners and turns the diffusion-specific quality-speed trade-off into a deterministic, predictable speedup set by the block size $B$, yielding up to $5.8\times$ wall-clock speedup over token-by-token MDLM decoding without modifying the underlying denoiser. Applied as a drop-in post-filter, dilated spacing also improves adaptive samplers. Code is available at https://github.com/omerlux/DUS.

cs.CL↗

arXiv2025

HeavyWater and SimplexWater: Distortion-Free LLM Watermarks for Low-Entropy Next-Token Predictions

Large language model (LLM) watermarks enable authentication of text provenance, curb misuse of machine-generated text, and promote trust in AI systems. Current watermarks operate by changing the next-token predictions output by an LLM. The updated (i.e., watermarked) predictions depend on random side information produced, for example, by hashing previously generated tokens. LLM watermarking is particularly challenging in low-entropy generation tasks -- such as coding -- where next-token predictions are near-deterministic. In this paper, we propose an optimization framework for watermark design. Our goal is to understand how to most effectively use random side information in order to maximize the likelihood of watermark detection and minimize the distortion of generated text. Our analysis informs the design of two new watermarks: HeavyWater and SimplexWater. Both watermarks are tunable, gracefully trading-off between detection accuracy and text distortion. They can also be applied to any LLM and are agnostic to side information generation. We examine the performance of HeavyWater and SimplexWater through several benchmarks, demonstrating that they can achieve high watermark detection accuracy with minimal compromise of text generation quality, particularly in the low-entropy regime. Our theoretical analysis also reveals surprising new connections between LLM watermarking and coding theory. The code implementation can be found in https://github.com/DorTsur/HeavyWater_SimplexWater

cs.CR↗

arXiv2025

Neural Estimation for Scaling Entropic Multimarginal Optimal Transport

Multimarginal optimal transport (MOT) is a powerful framework for modeling interactions between multiple distributions, yet its applicability is bottlenecked by a high computational overhead. Entropic regularization provides computational speedups via the multimarginal Sinkhorn algorithm, whose time complexity, for a dataset size $n$ and $k$ marginals, generally scales as $O(n^k)$. However, this dependence on the dataset size $n$ is computationally prohibitive for many machine learning problems. In this work, we propose a new computational framework for entropic MOT, dubbed Neural Entropic MOT (NEMOT), that enjoys significantly improved scalability. NEMOT employs neural networks trained using mini-batches, which transfers the computational complexity from the dataset size to the size of the mini-batch, leading to substantial gains. We provide formal guarantees on the accuracy of NEMOT via non-asymptotic error bounds. We supplement these with numerical results that demonstrate the performance gains of NEMOT over Sinkhorn's algorithm, as well as extensions to neural computation of multimarginal entropic Gromov-Wasserstein alignment. In particular, orders-of-magnitude speedups are observed relative to the state-of-the-art, with a notable increase in the feasible number of samples and marginals. NEMOT seamlessly integrates as a module in large-scale machine learning pipelines, and can serve to expand the practical applicability of entropic MOT for tasks involving multimarginal data.

cs.LG↗

arXiv2025

Optimized Couplings for Watermarking Large Language Models

Large-language models (LLMs) are now able to produce text that is, in many cases, seemingly indistinguishable from human-generated content. This has fueled the development of watermarks that imprint a ``signal'' in LLM-generated text with minimal perturbation of an LLM's output. This paper provides an analysis of text watermarking in a one-shot setting. Through the lens of hypothesis testing with side information, we formulate and analyze the fundamental trade-off between watermark detection power and distortion in generated textual quality. We argue that a key component in watermark design is generating a coupling between the side information shared with the watermark detector and a random partition of the LLM vocabulary. Our analysis identifies the optimal coupling and randomization strategy under the worst-case LLM next-token distribution that satisfies a min-entropy constraint. We provide a closed-form expression of the resulting detection rate under the proposed scheme and quantify the cost in a max-min sense. Finally, we provide an array of numerical results, comparing the proposed scheme with the theoretical optimum and existing schemes, in both synthetic data and LLM watermarking. Our code is available at https://github.com/Carol-Long/CC_Watermark

cs.CR↗

arXiv2025

Efficient Time Series Forecasting via Hyper-Complex Models and Frequency Aggregation

Time series forecasting is a long-standing problem in statistics and machine learning. One of the key challenges is processing sequences with long-range dependencies. To that end, a recent line of work applied the short-time Fourier transform (STFT), which partitions the sequence into multiple subsequences and applies a Fourier transform to each separately. We propose the Frequency Information Aggregation (FIA)-Net, which is based on a novel complex-valued MLP architecture that aggregates adjacent window information in the frequency domain. To further increase the receptive field of the FIA-Net, we treat the set of windows as hyper-complex (HC) valued vectors and employ HC algebra to efficiently combine information from all STFT windows altogether. Using the HC-MLP backbone allows for improved handling of sequences with long-term dependence. Furthermore, due to the nature of HC operations, the HC-MLP uses up to three times fewer parameters than the equivalent standard window aggregation method. We evaluate the FIA-Net on various time-series benchmarks and show that the proposed methodologies outperform existing state of the art methods in terms of both accuracy and efficiency. Our code is publicly available on https://anonymous.4open.science/r/research-1803/.

cs.LG↗

arXiv2024

InfoMat: A Tool for the Analysis and Visualization Sequential Information Transfer

Despite the popularity of information measures in analysis of probabilistic systems, proper tools for their visualization are not common. This work develops a simple matrix representation of information transfer in sequential systems, termed information matrix (InfoMat). The simplicity of the InfoMat provides a new visual perspective on existing decomposition formulas of mutual information, and enables us to prove new relations between sequential information theoretic measures. We study various estimation schemes of the InfoMat, facilitating the visualization of information transfer in sequential datasets. By drawing a connection between visual patterns in the InfoMat and various dependence structures, we observe how information transfer evolves in the dataset. We then leverage this tool to visualize the effect of capacity-achieving coding schemes on the underlying exchange of information. We believe the InfoMat is applicable to any time-series task for a better understanding of the data at hand.

cs.IT↗

arXiv2023

Multi-Agent Reinforcement Learning for Network Routing in Integrated Access Backhaul Networks

We investigate the problem of wireless routing in integrated access backhaul (IAB) networks consisting of fiber-connected and wireless base stations and multiple users. The physical constraints of these networks prevent the use of a central controller, and base stations have limited access to real-time network conditions. We aim to maximize packet arrival ratio while minimizing their latency, for this purpose, we formulate the problem as a multi-agent partially observed Markov decision process (POMDP). To solve this problem, we develop a Relational Advantage Actor Critic (Relational A2C) algorithm that uses Multi-Agent Reinforcement Learning (MARL) and information about similar destinations to derive a joint routing policy on a distributed basis. We present three training paradigms for this algorithm and demonstrate its ability to achieve near-centralized performance. Our results show that Relational A2C outperforms other reinforcement learning algorithms, leading to increased network efficiency and reduced selfish agent behavior. To the best of our knowledge, this work is the first to optimize routing strategy for IAB networks.

cs.NI↗

arXiv2023

Data-Driven Optimization of Directed Information over Discrete Alphabets

Directed information (DI) is a fundamental measure for the study and analysis of sequential stochastic models. In particular, when optimized over input distributions it characterizes the capacity of general communication channels. However, analytic computation of DI is typically intractable and existing optimization techniques over discrete input alphabets require knowledge of the channel model, which renders them inapplicable when only samples are available. To overcome these limitations, we propose a novel estimation-optimization framework for DI over discrete input spaces. We formulate DI optimization as a Markov decision process and leverage reinforcement learning techniques to optimize a deep generative model of the input process probability mass function (PMF). Combining this optimizer with the recently developed DI neural estimator, we obtain an end-to-end estimation-optimization algorithm which is applied to estimating the (feedforward and feedback) capacity of various discrete channels with memory. Furthermore, we demonstrate how to use the optimized PMF model to (i) obtain theoretical bounds on the feedback capacity of unifilar finite-state channels; and (ii) perform probabilistic shaping of constellations in the peak power-constrained additive white Gaussian noise channel.

cs.IT↗

arXiv2022

Neural Estimation and Optimization of Directed Information over Continuous Spaces

This work develops a new method for estimating and optimizing the directed information rate between two jointly stationary and ergodic stochastic processes. Building upon recent advances in machine learning, we propose a recurrent neural network (RNN)-based estimator which is optimized via gradient ascent over the RNN parameters. The estimator does not require prior knowledge of the underlying joint and marginal distributions. The estimator is also readily optimized over continuous input processes realized by a deep generative model. We prove consistency of the proposed estimation and optimization methods and combine them to obtain end-to-end performance guarantees. Applications for channel capacity estimation of continuous channels with memory are explored, and empirical results demonstrating the scalability and accuracy of our method are provided. When the channel is memoryless, we investigate the mapping learned by the optimized input generator.

cs.IT↗

arXiv2022

The Feedback Capacity of Noisy Output is the STate (NOST) Channels

We consider finite-state channels (FSCs) where the channel state is stochastically dependent on the previous channel output. We refer to these as Noisy Output is the STate (NOST) channels. We derive the feedback capacity of NOST channels in two scenarios: with and without causal state information (CSI) available at the encoder. If CSI is unavailable, the feedback capacity is $C_{\text{FB}}= \max_{P(x|y')} I(X;Y|Y')$, while if it is available at the encoder, the feedback capacity is $C_{\text{FB-CSI}}= \max_{P(u|y'),x(u,s')} I(U;Y|Y')$, where $U$ is an auxiliary RV with finite cardinality. In both formulas, the output process is a Markov process with stationary distribution. The derived formulas generalize special known instances from the literature, such as where the state is i.i.d. and where it is a deterministic function of the output. $C_{\text{FB}}$ and $C_{\text{FB-CSI}}$ are also shown to be computable via convex optimization problem formulations. Finally, we present an example of an interesting NOST channel for which CSI available at the encoder does not increase the feedback capacity.

cs.IT↗

arXiv2021

A Study On Data Augmentation In Voice Anti-Spoofing

In this paper, we perform an in-depth study of how data augmentation techniques improve synthetic or spoofed audio detection. Specifically, we propose methods to deal with channel variability, different audio compressions, different band-widths, and unseen spoofing attacks, which have all been shown to significantly degrade the performance of audio-based systems and Anti-Spoofing systems. Our results are based on the ASVspoof 2021 challenge, in the Logical Access (LA) and Deep Fake (DF) categories. Our study is Data-Centric, meaning that the models are fixed and we significantly improve the results by making changes in the data. We introduce two forms of data augmentation - compression augmentation for the DF part, compression & channel augmentation for the LA part. In addition, a new type of online data augmentation, SpecAverage, is introduced in which the audio features are masked with their average value in order to improve generalization. Furthermore, we introduce a Log spectrogram feature design that improved the results. Our best single system and fusion scheme both achieve state-of-the-art performance in the DF category, with an EER of 15.46% and 14.46% respectively. Our best system for the LA task reduced the best baseline EER by 50% and the min t-DCF by 16%. Our techniques to deal with spoofed data from a wide variety of distributions can be replicated and can help anti-spoofing and speech-based systems enhance their results.

cs.SD↗