arXiv subjects

Dhruv Jain

Publications and source records attributed to Dhruv Jain.

At least 19 recordsLinked to original sources

arXiv2026

Reason, Reward, Refine: Step-Level Errors Corrections with Structured Feedback for Physics Reasoning in Small Language Models

Physics reasoning fails structurally in small language models: an error at any step propagates forward, corrupting every inference that follows. Limited domain knowledge, hallucination under multi-step derivation, and distributional sensitivity compound this failure. We propose a step-level reward framework that identifies the first reasoning error, generates targeted structured feedback, and trains the model to revise its solution via policy gradient with KL regularization, without exposing it to ground truth solutions as generation targets. Unlike annotation-dependent step-level methods, no preference data construction is required and the external verifier operates exclusively at training time. Across five physics benchmarks, our framework delivers accuracy gains of 17-20% over CoT prompting and 10-16% over the strongest baseline, reduces calculation errors from 56.9% to 23.5%, and reduces miscomprehension errors from 22.3% to 12.0% in the best observed cases. Conceptual errors reduce from 89.7% to 68.7%, yet persist as the hardest failure mode across all conditions.

cs.AI↗

arXiv2026

Unmixing The Crowd: Learning Persistent Speaker Representations from Mixture-Derived Multi-Speaker Embeddings

We study whether persistent conversational speaker structure can be extracted directly from local overlapping speech mixtures. We propose a teacher-student framework that learns mixture-derived multi-speaker embeddings using only short overlapping segments and permutation-invariant latent supervision. Despite never being explicitly trained for speaker tracking, diarization, or conversational memory, the learned embedding space supports long-form speaker re-identification when combined with a lightweight online memory mechanism during inference. We additionally observe that the learned representation retains meaningful speaker structure under unseen overlap cardinalities. We further show that embeddings extracted from separation-first pipelines exhibit degraded clustering structure compared to embeddings predicted directly from mixtures. Finally, the learned embeddings remain effective for the downstream target speaker extraction task across multiple architectures. These findings suggest that local mixture-derived representations support persistent conversational speaker re-identification when combined with lightweight inference-time memory consolidation.

eess.AS↗

arXiv2026

Sona: Real-Time Multi-Target Sound Attenuation for Noise Sensitivity

For people with noise sensitivity, everyday soundscapes can be overwhelming. Existing tools such as active noise cancellation reduce discomfort by suppressing the entire acoustic environment, often at the cost of awareness of surrounding people and events. We present Sona, an interactive mobile system for real-time soundscape mediation that selectively attenuates bothersome sounds while preserving desired audio. Sona is built on a target-conditioned neural pipeline that supports simultaneous attenuation of multiple overlapping sound sources, overcoming the single-target limitation of prior systems. It runs in real time on-device and supports user-extensible sound classes through in-situ audio examples, without retraining. Sona is informed by a formative study with 68 noise-sensitive individuals. Through technical benchmarking and an in-situ study with 10 participants, we show that Sona achieves low-latency, multi-target attenuation suitable for live listening, and enables meaningful reductions in bothersome sounds while maintaining awareness of surroundings. These results point toward a new class of personal AI systems that support comfort and social participation by mediating real-world acoustic environments.

cs.SD↗

arXiv2026

VoiceAgentBench: Are Voice Assistants ready for agentic tasks?

Large scale Speech Language Models have enabled voice assistants capable of understanding natural spoken queries and performing complex tasks. However, existing speech benchmarks largely focus on isolated capabilities such as transcription or question answering and do not systematically evaluate agentic behavior or adversarial robustness. To address this, we introduce VoiceAgentBench, a comprehensive benchmark for evaluating SpeechLMs in realistic spoken agentic settings, comprising 6,000+ synthetic spoken queries spanning single-tool invocations, multi-tool workflows, multi-turn dialogue, and safety evaluations across English and six Indic languages. To ensure speaker diversity, we further simulate speaker variability using a novel sampling strategy that selects audios for TTS voice conversion based on speaker embeddings to maximize acoustic diversity. Our evaluation measures tool selection accuracy, structural consistency, and the correctness of tool invocations, including adversarial robustness. Across agentic tasks, ASR-LLM pipelines outperform end-to-end SpeechLMs, achieving up to 60.6% average parameter-filling accuracy on English, while SpeechLMs exhibit lower performance and sharper degradation on Indic languages. All models struggle in sequential workflows and safety evaluations, highlighting persistent limitations in tool orchestration, multilingual generalization, and safety robustness. VoiceAgentBench is publicly available on Hugging Face at https://huggingface.co/datasets/krutrim-ai-labs/VoiceAgentBench, and the codebase is released at https://github.com/ola-krutrim/VoiceAgentBench.

cs.AI↗

arXiv2026

RAVEN: Realtime Accessibility in Virtual ENvironments for Blind and Low-Vision People

As virtual 3D environments become more prevalent, equitable access is essential for blind and low-vision (BLV) users, who face challenges with spatial awareness, navigation, and interaction. Prior work has explored supplementing visual information with auditory or haptic modalities, but these methods are static and offer limited support for dynamic, in-context adaptation. Recent advances in generative AI allow users to query and modify 3D scenes via natural language, introducing a paradigm that offers greater flexibility and control for accessibility. We present RAVEN, a system that enables BLV users to issue queries and modification prompts to improve the runtime accessibility of 3D virtual scenes. We evaluated RAVEN with eight BLV people and six Unity developers, generating empirical insights into how conversational programming can support personalized accessibility in 3D environments. Our work highlights both the promise of natural language interaction-intuitive, flexible, and empowering-and the challenges of ensuring reliability, transparency, and trust in generative AI-driven accessibility systems.

cs.HC↗

arXiv2025

SN 2024aecx: a fast-evolving Type IIb supernova with a prominent shock-cooling peak

SN 2024aecx is a nearby ($\sim$11 Mpc) Type IIb SN discovered within $\sim$1 d after explosion. In this paper we report high-cadence photometric (typically 0.5$\sim$1 day) and spectroscopic follow-up observations, conducted from as early as 0.27 d post discovery out to the nebular phase at 158.4 d. We analyze the environment of SN 2024aecx and derive a new distance (11.3$\pm$1.1 Mpc), metallicity and host extinction. The light curve exhibits a hot and luminous shock-cooling peak at the first few days, followed by a main peak with very rapid post-maximum decline. The earliest spectra are blue and featureless, while from 2.3 d after discovery prominent P-Cygni profiles emerge. At nebular phase, the emission lines exhibit asymmetric and double-peaked profiles, indicating asphericity and/or early dust formation in the ejecta. Nebular spectral modelling indicates a blueshifted O-rich clump moving toward observer, and the $[\text{OI}]/[\text{CaII}]$ line ratio suggests an intermediate-mass progenitor. We simulated the progenitor and explosion using a two-component model of shock cooling and radioactive $^{56}$Ni heating; our model favors an extended, low-mass H-rich envelope with $M_{\mathrm{e}} = 0.04\pm{0.01} M_{\odot}$ and a low ejecta mass of$M_{\mathrm{ej}} = 1.55^{+0.18}_{-0.14} M_{\odot}$. And the nebular-phase spectra and light-curve modelling both suggest that it most likely originated from an intermediate-mass binary progenitor system. The comprehensive monitoring of SN 2024aecx, coupled with the detailed characterization of its local environment, establishes it as a benchmark event for probing the progenitors and explosion mechanisms of Type IIb SNe.

astro-ph.SR↗

arXiv2025

PERCS: Persona-Guided Controllable Biomedical Summarization Dataset

Automatic medical text simplification plays a key role in improving health literacy by making complex biomedical research accessible to diverse readers. However, most existing resources assume a single generic audience, overlooking the wide variation in medical literacy and information needs across user groups. To address this limitation, we introduce PERCS (Persona-guided Controllable Summarization), a dataset of biomedical abstracts paired with summaries tailored to four personas: Laypersons, Premedical Students, Non-medical Researchers, and Medical Experts. These personas represent different levels of medical literacy and information needs, emphasizing the need for targeted, audience-specific summarization. Each summary in PERCS was reviewed by physicians for factual accuracy and persona alignment using a detailed error taxonomy. Technical validation shows clear differences in readability, vocabulary, and content depth across personas. Along with describing the dataset, we benchmark four large language models on PERCS using automatic evaluation metrics that assess comprehensiveness, readability, and faithfulness, establishing baseline results for future research. The dataset, annotation guidelines, and evaluation materials are publicly available to support research on persona-specific communication and controllable biomedical summarization.

cs.CL↗

arXiv2025

What do gravitational-wave observations tell us about Luminous Red Novae?

Luminous Red Novae (LRNe) have been argued to be related to the ejection of common envelopes (CEs) in binary star systems. Ejection of CEs leads to tightened stellar orbits capable of forming compact binaries that merge in Hubble time. As these mergers are seen by gravitational-wave (GW) detectors such as LIGO, Virgo and KAGRA (LVK), we ask what the merger rates of compact binaries in LVK tell us about the fraction of LRNe that lead to the formation of compact binaries that merge in Hubble time. Using the observed volumetric rates of LRNe from the Zwicky Transient Facility (ZTF) and of compact binary mergers from LVK observations, we derive limits on the fraction of LRNe that produce compact binaries that merge in Hubble time. Assuming the LRNe rate closely follows the star formation rate at any redshift, we use the delay time distribution models for compact binaries to compute the compact binary merger rate. A comparison of this merger rate with the latest volumetric rates of compact binary mergers from the fourth GW transient catalog (GWTC-4) at the present epoch of LVK allows us to constrain the above fraction. We find that only a fraction as small as $\sim 10^{-3}$ (median) of the LRNe correspond to the GW-observed binary neutron star (BNS) and neutron star-black hole (NSBH) mergers. This potentially implies that the majority of the LRNe population will not lead to mergers of compact objects, but other end products, such as stellar mergers.

astro-ph.HE↗

arXiv2025

EvolveCaptions: Empowering DHH Users Through Real-Time Collaborative Captioning

Automatic Speech Recognition (ASR) systems often fail to accurately transcribe speech from Deaf and Hard of Hearing (DHH) individuals, especially during real-time conversations. Existing personalization approaches typically require extensive pre-recorded data and place the burden of adaptation on the DHH speaker. We present EvolveCaptions, a real-time, collaborative ASR adaptation system that supports in-situ personalization with minimal effort. Hearing participants correct ASR errors during live conversations. Based on these corrections, the system generates short, phonetically targeted prompts for the DHH speaker to record, which are then used to fine-tune the ASR model. In a study with 12 DHH and six hearing participants, EvolveCaptions reduced Word Error Rate (WER) across all DHH users within one hour of use, using only five minutes of recording time on average. Participants described the system as intuitive, low-effort, and well-integrated into communication. These findings demonstrate the promise of collaborative, real-time ASR adaptation for more equitable communication.

cs.HC↗

arXiv2025

RISE: Adaptive music playback for Realtime Intensity Synchronization with Exercise

We propose a system to adapt a user's music to their exercise by aligning high-energy music segments with intense intervals of the workout. Listening to music during exercise can boost motivation and performance. However, the structure of the music may be different from the user's natural phases of rest and work, causing users to rest longer than needed while waiting for a motivational section, or lose motivation mid-work if the section ends too soon. To address this, our system, called RISE, automatically estimates the intense segments in music and uses component-based music rearrangement techniques to dynamically extend and shorten different segments of the user's song to fit the ongoing exercise routine. Our system takes as input the rest and work durations to guide adaptation. Currently, this is determined either via a pre-defined plan or manual input during the workout. We evaluated RISE with 12 participants and compared our system to a non-adaptive music baseline while exercising in our lab. Participants found our rearrangements keeps intensity estimation accurate, and many recalled moments when intensity alignment helped them push through their workout.

cs.SD↗

arXiv2025

CapTune: Adapting Non-Speech Captions With Anchored Generative Models

Non-speech captions are essential to the video experience of deaf and hard of hearing (DHH) viewers, yet conventional approaches often overlook the diversity of their preferences. We present CapTune, a system that enables customization of non-speech captions based on DHH viewers' needs while preserving creator intent. CapTune allows caption authors to define safe transformation spaces using concrete examples and empowers viewers to personalize captions across four dimensions: level of detail, expressiveness, sound representation method, and genre alignment. Evaluations with seven caption creators and twelve DHH participants showed that CapTune supported creators' creative control while enhancing viewers' emotional engagement with content. Our findings also reveal trade-offs between information richness and cognitive load, tensions between interpretive and descriptive representations of sound, and the context-dependent nature of caption preferences.

cs.HC↗

arXiv2025

SonoCraftAR: Towards Supporting Personalized Authoring of Sound-Reactive AR Interfaces by Deaf and Hard of Hearing Users

Augmented reality (AR) has shown promise for supporting Deaf and hard-of-hearing (DHH) individuals by captioning speech and visualizing environmental sounds, yet existing systems do not allow users to create personalized sound visualizations. We present SonoCraftAR, a proof-of-concept prototype that empowers DHH users to author custom sound-reactive AR interfaces using typed natural language input. SonoCraftAR integrates real-time audio signal processing with a multi-agent LLM pipeline that procedurally generates animated 2D interfaces via a vector graphics library. The system extracts the dominant frequency of incoming audio and maps it to visual properties such as size and color, making the visualizations respond dynamically to sound. This early exploration demonstrates the feasibility of open-ended sound-reactive AR interface authoring and discusses future opportunities for personalized, AI-assisted tools to improve sound accessibility.

cs.HC↗

arXiv2025

Differential-UMamba: Rethinking Tumor Segmentation Under Limited Data Scenarios

In data-scarce scenarios, deep learning models often overfit to noise and irrelevant patterns, which limits their ability to generalize to unseen samples. To address these challenges in medical image segmentation, we introduce Diff-UMamba, a novel architecture that combines the UNet framework with the mamba mechanism to model long-range dependencies. At the heart of Diff-UMamba is a noise reduction module, which employs a signal differencing strategy to suppress noisy or irrelevant activations within the encoder. This encourages the model to filter out spurious features and enhance task-relevant representations, thereby improving its focus on clinically significant regions. As a result, the architecture achieves improved segmentation accuracy and robustness, particularly in low-data settings. Diff-UMamba is evaluated on multiple public datasets, including medical segmentation decathalon dataset (lung and pancreas) and AIIB23, demonstrating consistent performance gains of 1-3% over baseline methods in various segmentation tasks. To further assess performance under limited data conditions, additional experiments are conducted on the BraTS-21 dataset by varying the proportion of available training samples. The approach is also validated on a small internal non-small cell lung cancer dataset for the segmentation of gross tumor volume in cone beam CT, where it achieves a 4-5% improvement over baseline.

cs.CV↗

arXiv2025

Predicting Patient Survival with Airway Biomarkers using nn-Unet/Radiomics

The primary objective of the AIIB 2023 competition is to evaluate the predictive significance of airway-related imaging biomarkers in determining the survival outcomes of patients with lung fibrosis.This study introduces a comprehensive three-stage approach. Initially, a segmentation network, namely nn-Unet, is employed to delineate the airway's structural boundaries. Subsequently, key features are extracted from the radiomic images centered around the trachea and an enclosing bounding box around the airway. This step is motivated by the potential presence of critical survival-related insights within the tracheal region as well as pertinent information encoded in the structure and dimensions of the airway. Lastly, radiomic features obtained from the segmented areas are integrated into an SVM classifier. We could obtain an overall-score of 0.8601 for the segmentation in Task 1 while 0.7346 for the classification in Task 2.

cs.CV↗

arXiv2025

Knowledge Graphs are all you need: Leveraging KGs in Physics Question Answering

This study explores the effectiveness of using knowledge graphs generated by large language models to decompose high school-level physics questions into sub-questions. We introduce a pipeline aimed at enhancing model response quality for Question Answering tasks. By employing LLMs to construct knowledge graphs that capture the internal logic of the questions, these graphs then guide the generation of subquestions. We hypothesize that this method yields sub-questions that are more logically consistent with the original questions compared to traditional decomposition techniques. Our results show that sub-questions derived from knowledge graphs exhibit significantly improved fidelity to the original question's logic. This approach not only enhances the learning experience by providing clearer and more contextually appropriate sub-questions but also highlights the potential of LLMs to transform educational methodologies. The findings indicate a promising direction for applying AI to improve the quality and effectiveness of educational content.

cs.CL↗

arXiv2024

Improving Physics Reasoning in Large Language Models Using Mixture of Refinement Agents

Large Language Models (LLMs) demonstrate remarkable capabilities in various reasoning tasks. However, they encounter significant challenges when it comes to scientific reasoning, particularly in physics, which requires not only mathematical reasoning but also factual and conceptual understanding. When addressing complex physics problems, LLMs typically face three key issues: problem miscomprehension, incorrect concept application, and computational errors. While each of these problems can be addressed individually, there is a need for a generalized approach that can tackle all three issues simultaneously. To address this, we introduce Mixture of Refinement Agents (MoRA), a novel agentic refinement framework that iteratively refines the LLM generated base solution by correcting the aforementioned errors, resulting in a significant performance improvement for open-source LLMs. Our approach aims to bridge the gap between opensource LLMs and GPT-4o by utilizing the latter as error identifier to guide these refinement agents. We evaluate our approach on the SciEval and MMLU subsets along with our own physics dataset (PhysicsQA). MoRA significantly improves the performance of Llama-3-70B and Gemma-2-27B on these datasets, achieving up to a 16% increase in final answer accuracy.

cs.AI↗

arXiv2024

Rates and beaming angles of GRBs associated with compact binary coalescences

Some, if not all, binary neutron star (BNS) coalescences, and a fraction of neutron - star black hole (NSBH) mergers, are thought to produce sufficient mass-ejection to power Gamma-Ray Bursts (GRBs). However, this fraction, as well as the distribution of beaming angles of BNS-associated GRBs, are poorly constrained from observation. Recent work applied machine learning tools to analyze GRB light curves observed by {\textit{Fermi}}/GBM and {\it Swift}/BAT. GRBs were segregated into multiple distinct clusters, with the tantalizing possibility that one of them (BNS cluster) could be associated with BNSs and another (NSBH cluster) with NSBHs. As a proof of principle, assuming that all GRBs detected by {\it Fermi}/GBM and {\it Swift}/BAT associated with BNSs (NSBHs) lie in the BNS (NSBH) cluster, we estimate their rates ($\mathrm{Gpc}^{-3}\mathrm{yr}^{-1}$). We compare these rates with corresponding BNS and NSBH rates estimated by the LIGO-Virgo-Kagra (LVK) collaboration from the first three observing runs (O1, O2, O3). We find that the BNS rates are consistent with LVK's rate estimates, assuming a uniform distribution of beaming fractions ($f_b \in [0.01, 0.1]$). Conversely, using the LVK's BNS rate estimates, assuming all BNS mergers produce GRBs, we are able to constrain the beaming angle distribution to $θ_j \in [0.8^{\circ}, 33.5^{\circ}]$ at $90\%$ confidence. We similarly place limits on the fraction of GRB-Bright NSBHs as $f_B \in [1.3\%, 63\%]$ ($f_B \in [0.4\%, 15\%]$) with {\it Fermi}/GBM ({\it Swift}/BAT) data.

astro-ph.HE↗

arXiv2024

SoundShift: Exploring Sound Manipulations for Accessible Mixed-Reality Awareness

Mixed-reality (MR) soundscapes blend real-world sound with virtual audio from hearing devices, presenting intricate auditory information that is hard to discern and differentiate. This is particularly challenging for blind or visually impaired individuals, who rely on sounds and descriptions in their everyday lives. To understand how complex audio information is consumed, we analyzed online forum posts within the blind community, identifying prevailing challenges, needs, and desired solutions. We synthesized the results and propose SoundShift for increasing MR sound awareness, which includes six sound manipulations: Transparency Shift, Envelope Shift, Position Shift, Style Shift, Time Shift, and Sound Append. To evaluate the effectiveness of SoundShift, we conducted a user study with 18 blind participants across three simulated MR scenarios, where participants identified specific sounds within intricate soundscapes. We found that SoundShift increased MR sound awareness and minimized cognitive load. Finally, we developed three real-world example applications to demonstrate the practicality of SoundShift.

cs.HC↗