Vipul Arora

Sumit Kumar, B. Anshuman, Linus Ruettimann et al. · eth-zurich

Event detection improves when events are captured by two different modalities rather than just one. But to train detection systems on multiple modalities is challenging, in particular when there is abundance of unlabelled data but limited amounts of labeled data. We develop a novel self-supervised learning technique for multi-modal data that learns (hidden) correlations between simultaneously recorded microphone (sound) signals and accelerometer (body vibration) signals. The key objective of this work is to learn useful embeddings associated with high performance in downstream event detection tasks when labeled data is scarce and the audio events of interest (songbird vocalizations) are sparse. We base our approach on deep canonical correlation analysis (DCCA) that suffers from event sparseness. We overcome the sparseness of positive labels by first learning a data sampling model from the labelled data and by applying DCCA on the output it produces. This method that we term balanced DCCA (b-DCCA) improves the performance of the unsupervised embeddings on the downstream supervised audio detection task compared to classsical DCCA. Because data labels are frequently imbalanced, our method might be of broad utility in low-resource scenarios.

Anup Singh, Kris Demuynck, Vipul Arora

Audio fingerprinting systems must efficiently and robustly identify query snippets in an extensive database. To this end, state-of-the-art systems use deep learning to generate compact audio fingerprints. These systems deploy indexing methods, which quantize fingerprints to hash codes in an unsupervised manner to expedite the search. However, these methods generate imbalanced hash codes, leading to their suboptimal performance. Therefore, we propose a self-supervised learning framework to compute fingerprints and balanced hash codes in an end-to-end manner to achieve both fast and accurate retrieval performance. We model hash codes as a balanced clustering process, which we regard as an instance of the optimal transport problem. Experimental results indicate that the proposed approach improves retrieval efficiency while preserving high accuracy, particularly at high distortion levels, compared to the competing methods. Moreover, our system is efficient and scalable in computational load and memory storage.

Vipul Arora, Arnab Bhattacharyya, Clément L. Canonne et al.

This paper considers the problem of testing the maximum in-degree of the Bayes net underlying an unknown probability distribution $P$ over $\{0,1\}^n$, given sample access to $P$. We show that the sample complexity of the problem is $\tildeΘ(2^{n/2}/\varepsilon^2)$. Our algorithm relies on a testing-by-learning framework, previously used to obtain sample-optimal testers; in order to apply this framework, we develop new algorithms for ``near-proper'' learning of Bayes nets, and high-probability learning under $χ^2$ divergence, which are of independent interest.

Swapnil Dey, Vipul Arora, Sachchida Nand Tripathi

Air quality monitoring is becoming an essential task with rising awareness about air quality. Low cost air quality sensors are easy to deploy but are not as reliable as the costly and bulky reference monitors. The low quality sensors can be calibrated against the reference monitors with the help of deep learning. In this paper, we translate the task of sensor calibration into a semi-supervised domain adaptation problem and propose a novel solution for the same. The problem is challenging because it is a regression problem with covariate shift and label gap. We use histogram loss instead of mean squared or mean absolute error, which is commonly used for regression, and find it useful against covariate shift. To handle the label gap, we propose weighting of samples for adversarial entropy optimization. In experimental evaluations, the proposed scheme outperforms many competitive baselines, which are based on semi-supervised and supervised domain adaptation, in terms of R2 score and mean absolute error. Ablation studies show the relevance of each proposed component in the entire scheme.

Vipul Arora, Arnab Bhattacharyya, Philips George John et al.

Over the last three decades, function testing has been extensively studied over Boolean, finite fields, and discrete settings. However, to encode the real-world applications more succinctly, function testing over the reals (where the domain and range, both are reals) is of prime importance. Recently, there have been some works in the direction of testing for algebraic representations of such functions: the work by Fleming and Yoshida (ITCS 20), Arora, Kelman, and Meir (SOSA 25) on linearity testing and the work of Arora, Bhattacharyya, Fleming, Kelman, and Yoshida (SODA 23) for testing low-degree polynomials. Our work follows the same avenue, wherein we study three well-studied sparse representations of functions, over the reals, namely (i) $k$-linearity, (ii) $k$-sparse polynomials, and (iii) $k$-junta. In this setting, given approximate query access to some $f:\mathbb{R}^n \rightarrow \mathbb{R}$, we want to decide if the function satisfies some property of interest, or if it is far from all functions that satisfy the property. Here, the distance is measured in the $\ell_1$-metric, under the assumption that we are drawing samples from the Standard Gaussian distribution. We present efficient testers and $Ω(k)$ lower bounds for testing each of these three properties.

Sumit Kumar, Suraj Jaiswal, Parampreet Singh et al.

The advancement of machine learning in audio analysis has opened new possibilities for technology-enhanced music education. This paper introduces a framework for automatic singing mistake detection in the context of music pedagogy, supported by a newly curated dataset. The dataset comprises synchronized teacher learner vocal recordings, with annotations marking different types of mistakes made by learners. Using this dataset, we develop different deep learning models for mistake detection and benchmark them. To compare the efficacy of mistake detection systems, a new evaluation methodology is proposed. Experiments indicate that the proposed learning-based methods are superior to rule-based methods. A systematic study of errors and a cross-teacher study reveal insights into music pedagogy that can be utilised for various music applications. This work sets out new directions of research in music pedagogy. The codes and dataset are publicly available.

Adhiraj Banerjee, Vipul Arora

Many operations on sensory data -- comparison, memory, retrieval, and reasoning -- are naturally expressed over discrete symbolic structures. In language this interface is given by tokens; in audio, it must be learned. Existing audio tokenizers rely on quantization, clustering, or codec reconstruction, assigning tokens locally, so sequence consistency, compactness, length control, termination, and edit similarity are rarely optimized directly. We introduce PairAlign, a framework for compact audio tokenization through sequence-level self-alignment. PairAlign treats tokenization as conditional sequence generation: an encoder maps speech to a continuous condition, and an autoregressive decoder generates tokens from BOS, learning token identity, order, length, and EOS placement. Given two content-preserving views, each view's sequence is trained to be likely under the other's representation, while unrelated examples provide competing sequences. This gives a scalable surrogate for edit-distance preservation while discouraging many-to-one collapse. PairAlign starts from VQ-style tokenization and refines it with EMA-teacher targets, cross-paired teacher forcing, prefix corruption, likelihood contrast, and length control. On 3-second speech, PairAlign learns compact, non-degenerate sequences with broad vocabulary usage and strong cross-view consistency. On TIMIT retrieval, it preserves edit-distance search while reducing archive token count by 55%. A continuous-sweep probe shows lower local overlap than a dense geometric tokenizer, but stronger length control and bounded edit trajectories under 100 ms shifts. PairAlign is a sequence-symbolic predictive learner: like JEPA-style objectives, it predicts an abstract target from another view as a learned variable-length symbolic sequence, not a continuous latent.

Parampreet Singh, Akshay Raina, Sayeedul Islam Sheikh et al.

Supervised machine learning frameworks rely on extensive labeled datasets for robust performance on real-world tasks. However, there is a lack of large annotated datasets in audio and music domains, as annotating such recordings is resource-intensive, laborious, and often require expert domain knowledge. In this work, we explore the use of label propagation (LP), a graph-based semi-supervised learning technique, for automatically labeling the unlabeled set in an unsupervised manner. By constructing a similarity graph over audio embeddings, we propagate limited label information from a small annotated subset to a larger unlabeled corpus in a transductive, semi-supervised setting. We apply this method to two tasks in Indian Art Music (IAM): Raga identification and Instrument classification. For both these tasks, we integrate multiple public datasets along with additional recordings we acquire from Prasar Bharati Archives to perform LP. Our experiments demonstrate that LP significantly reduces labeling overhead and produces higher-quality annotations compared to conventional baseline methods, including those based on pretrained inductive models. These results highlight the potential of graph-based semi-supervised learning to democratize data annotation and accelerate progress in music information retrieval.

Parampreet Singh, Somya Kumar, Chaitanya Shailendra Nitawe et al.

Raga identification in Indian Art Music (IAM) remains challenging due to the presence of numerous rarely performed Ragas that are not represented in available training datasets. Traditional classification models struggle in this setting, as they assume a closed set of known categories and therefore fail to recognise or meaningfully group previously unseen Ragas. Recent works have tried categorizing unseen Ragas, but they run into a problem of catastrophic forgetting, where the knowledge of previously seen Ragas is diminished. To address this problem, we adopt a unified learning framework that leverages both labeled and unlabeled audio, enabling the model to discover coherent categories corresponding to the unseen Ragas, while retaining the knowledge of previously known ones. We test our model on benchmark Raga Identification datasets and demonstrate its performance in categorizing previously seen, unseen, and all Raga classes. The proposed approach surpasses the previous NCD-based pipeline even in discovering the unseen Raga categories, offering new insights into representation learning for IAM tasks.

Nagarathna Ravi, Thishyan Raj T, Vipul Arora

Confidence estimation of predictions from an End-to-End (E2E) Automatic Speech Recognition (ASR) model benefits ASR's downstream and upstream tasks. Class-probability-based confidence scores do not accurately represent the quality of overconfident ASR predictions. An ancillary Confidence Estimation Model (CEM) calibrates the predictions. State-of-the-art (SOTA) solutions use binary target scores for CEM training. However, the binary labels do not reveal the granular information of predicted words, such as temporal alignment between reference and hypothesis and whether the predicted word is entirely incorrect or contains spelling errors. Addressing this issue, we propose a novel Temporal-Lexeme Similarity (TeLeS) confidence score to train CEM. To address the data imbalance of target scores while training CEM, we use shrinkage loss to focus on hard-to-learn data points and minimise the impact of easily learned data points. We conduct experiments with ASR models trained in three languages, namely Hindi, Tamil, and Kannada, with varying training data sizes. Experiments show that TeLeS generalises well across domains. To demonstrate the applicability of the proposed method, we formulate a TeLeS-based Acquisition (TeLeS-A) function for sampling uncertainty in active learning. We observe a significant reduction in the Word Error Rate (WER) as compared to SOTA methods.

Vikas Kanaujia, Mathias S. Scheurer, Vipul Arora

Deep generative models complement Markov-chain-Monte-Carlo methods for efficiently sampling from high-dimensional distributions. Among these methods, explicit generators, such as Normalising Flows (NFs), in combination with the Metropolis Hastings algorithm have been extensively applied to get unbiased samples from target distributions. We systematically study central problems in conditional NFs, such as high variance, mode collapse and data efficiency. We propose adversarial training for NFs to ameliorate these problems. Experiments are conducted with low-dimensional synthetic datasets and XY spin models in two spatial dimensions.

Saugat Kannojia, Anirudh Praveen, Danish Vasdev et al.

Dementia is a neurodegenerative disease that causes gradual cognitive impairment, which is very common in the world and undergoes a lot of research every year to prevent and cure it. It severely impacts the patient's ability to remember events and communicate clearly, where most variations of it have no known cure, but early detection can help alleviate symptoms before they become worse. One of the main symptoms of dementia is difficulty in expressing ideas through speech. This paper attempts to talk about a model developed to predict the onset of the disease using audio recordings from patients. An ASR-based model was developed that generates transcripts from the audio files using Whisper model and then applies RoBERTa regression model to generate an MMSE score for the patient. This score can be used to predict the extent to which the cognitive ability of a patient has been affected. We use the PROCESS_V1 dataset for this task, which is introduced through the PROCESS Grand Challenge 2025. The model achieved an RMSE score of 2.6911 which is around 10 percent lower than the described baseline.

Anup Singh, Kris Demuynck, Vipul Arora

Spoken term detection (STD) is often hindered by reliance on frame-level features and the computationally intensive DTW-based template matching, limiting its practicality. To address these challenges, we propose a novel approach that encodes speech into discrete, speaker-agnostic semantic tokens. This facilitates fast retrieval using text-based search algorithms and effectively handles out-of-vocabulary terms. Our approach focuses on generating consistent token sequences across varying utterances of the same term. We also propose a bidirectional state space modeling within the Mamba encoder, trained in a self-supervised learning framework, to learn contextual frame-level features that are further encoded into discrete tokens. Our analysis shows that our speech tokens exhibit greater speaker invariance than those from existing tokenizers, making them more suitable for STD tasks. Empirical evaluation on LibriSpeech and TIMIT databases indicates that our method outperforms existing STD baselines while being more efficient.

Vikas Kanaujia, Vipul Arora

Unnormalized probability distributions are central to modeling complex physical systems across various scientific domains. Traditional sampling methods, such as Markov Chain Monte Carlo (MCMC), often suffer from slow convergence, critical slowing down, poor mode mixing, and high autocorrelation. In contrast, likelihood-based and adversarial machine learning models, though effective, are heavily data-driven, requiring large datasets and often encountering mode covering and mode collapse. In this work, we propose ScoreNF, a score-based learning framework built on the Normalizing Flow (NF) architecture, integrated with an Independent Metropolis-Hastings (IMH) module, enabling efficient and unbiased sampling from unnormalized target distributions. We show that ScoreNF maintains high performance even with small training ensembles, thereby reducing reliance on computationally expensive MCMC-generated training data. We also present a method for assessing mode-covering and mode-collapse behaviours. We validate our method on synthetic 2D distributions (MOG-4 and MOG-8) and the high-dimensional $φ^4$ lattice field theory distribution, demonstrating its effectiveness for sampling tasks.

Adhiraj Banerjee, Vipul Arora

Text-guided sound separation supports flexible audio editing across media and assistive applications, but existing models like AudioSep are too compute-heavy for edge deployment. Neural audio codec (NAC) models such as CodecFormer and SDCodec are compute-efficient but limited to fixed-class separation. We introduce CodecSep, the first NAC-based model for on-device universal, text-driven separation. CodecSep combines DAC compression with a Transformer masker modulated by CLAP-derived FiLM parameters. Across six open-domain benchmarks under matched training/prompt protocols, \textbf{CodecSep} surpasses \textbf{AudioSep} in separation fidelity (SI-SDR) while remaining competitive in perceptual quality (ViSQOL) and matching or exceeding fixed-stem baselines (TDANet, CodecFormer, SDCodec). In code-stream deployments, it needs just 1.35~GMACs end-to-end -- approximately $54\times$ less compute ($25\times$ architecture-only) than spectrogram-domain separators like AudioSep -- while remaining fully bitstream-compatible.

Sumit Kumar, Parampreet Singh, Vipul Arora

Ornamentations, embellishments, or microtonal inflections are essential to melodic expression across many musical traditions, adding depth, nuance, and emotional impact to performances. Recognizing ornamentations in singing voices is key to MIR, with potential applications in music pedagogy, singer identification, genre classification, and controlled singing voice generation. However, the lack of annotated datasets and specialized modeling approaches remains a major obstacle for progress in this research area. In this work, we introduce Rāga Ornamentation Detection (ROD), a novel dataset comprising Indian classical music recordings curated by expert musicians. The dataset is annotated using a custom Human-in-the-Loop tool for six vocal ornaments marked as event-based labels. Using this dataset, we develop an ornamentation detection model based on deep time-series analysis, preserving ornament boundaries during the chunking of long audio recordings. We conduct experiments using different train-test configurations within the ROD dataset and also evaluate our approach on a separate, manually annotated dataset of Indian classical concert recordings. Our experimental results support the superior performance of our proposed approach over the baseline CRNN.

Parampreet Singh, Vipul Arora

Raga identification is an important problem within the domain of Indian Art music, as Ragas are fundamental to its composition and performance, playing a crucial role in music retrieval, preservation, and education. Few studies that have explored this task employ approaches such as signal processing, Machine Learning (ML), and more recently, Deep Learning (DL) based methods. However, a key question remains unanswered in all these works: do these ML/DL methods learn and interpret Ragas in a manner similar to human experts? Besides, a significant roadblock in this research is the unavailability of an ample supply of rich, labeled datasets, which drives these ML/DL-based methods. In this paper, firstly we curate a dataset comprising 191 hours of Hindustani Classical Music (HCM) recordings, annotate it for Raga and tonic labels, and train a CNN-LSTM model for the task of Automatic Raga Identification (ARI). We achieve a chunk-wise f1-measure of 0.89 for a subset of 12 Raga classes. Following this, we make one of the first attempts to employ model explainability techniques: SoundLIME and GradCAM++ for Raga identification, to evaluate whether the classifier's predictions align with human understanding of Ragas. We compare the generated explanations with human expert annotations and further analyze individual test examples to understand the role of regions highlighted by explanations in making correct or incorrect predictions made by the model. Our results demonstrate a significant alignment of the model's understanding with human understanding, and the thorough analysis validates the effectiveness of our approach.

Vipul Arora, Arnab Bhattacharyya, Mathews Boban et al.

We study the problem of robust multivariate polynomial regression: let $p\colon\mathbb{R}^n\to\mathbb{R}$ be an unknown $n$-variate polynomial of degree at most $d$ in each variable. We are given as input a set of random samples $(\mathbf{x}_i,y_i) \in [-1,1]^n \times \mathbb{R}$ that are noisy versions of $(\mathbf{x}_i,p(\mathbf{x}_i))$. More precisely, each $\mathbf{x}_i$ is sampled independently from some distribution $χ$ on $[-1,1]^n$, and for each $i$ independently, $y_i$ is arbitrary (i.e., an outlier) with probability at most $ρ< 1/2$, and otherwise satisfies $|y_i-p(\mathbf{x}_i)|\leqσ$. The goal is to output a polynomial $\hat{p}$, of degree at most $d$ in each variable, within an $\ell_\infty$-distance of at most $O(σ)$ from $p$. Kane, Karmalkar, and Price [FOCS'17] solved this problem for $n=1$. We generalize their results to the $n$-variate setting, showing an algorithm that achieves a sample complexity of $O_n(d^n\log d)$, where the hidden constant depends on $n$, if $χ$ is the $n$-dimensional Chebyshev distribution. The sample complexity is $O_n(d^{2n}\log d)$, if the samples are drawn from the uniform distribution instead. The approximation error is guaranteed to be at most $O(σ)$, and the run-time depends on $\log(1/σ)$. In the setting where each $\mathbf{x}_i$ and $y_i$ are known up to $N$ bits of precision, the run-time's dependence on $N$ is linear. We also show that our sample complexities are optimal in terms of $d^n$. Furthermore, we show that it is possible to have the run-time be independent of $1/σ$, at the cost of a higher sample complexity.

Kalpit Yadav, Vipul Arora, Sonu Kumar Jha et al.

Low-cost particulate matter sensors are transforming air quality monitoring because they have lower costs and greater mobility as compared to reference monitors. Calibration of these low-cost sensors requires training data from co-deployed reference monitors. Machine Learning based calibration gives better performance than conventional techniques, but requires a large amount of training data from the sensor, to be calibrated, co-deployed with a reference monitor. In this work, we propose novel transfer learning methods for quick calibration of sensors with minimal co-deployment with reference monitors. Transfer learning utilizes a large amount of data from other sensors along with a limited amount of data from the target sensor. Our extensive experimentation finds the proposed Model-Agnostic- Meta-Learning (MAML) based transfer learning method to be the most effective over other competitive baselines.

Shivam Pal, Vipul Arora, Pawan Goyal

A prerequisite is anything that you need to know or understand first before attempting to learn or understand something new. In the current work, we present a method of finding prerequisite relations between concepts using related textbooks. Previous researchers have focused on finding these relations using Wikipedia link structure through unsupervised and supervised learning approaches. In the current work, we have proposed two methods, one is statistical method and another is learning-based method. We mine the rich and structured knowledge available in the textbooks to find the content for those concepts and the order in which they are discussed. Using this information, proposed statistical method estimates explicit as well as implicit prerequisite relations between concepts. During experiments, we have found performance of proposed statistical method is better than the popular RefD method, which uses Wikipedia link structure. And proposed learning-based method has shown a significant increase in the efficiency of supervised learning method when compared with graph and text-based learning-based approaches.

Vipul Arora, Laxmidhar Behera, Ajay Pratap Yadav

A control theoretic approach is presented in this paper for both batch and instantaneous updates of weights in feed-forward neural networks. The popular Hamilton-Jacobi-Bellman (HJB) equation has been used to generate an optimal weight update law. The remarkable contribution in this paper is that closed form solutions for both optimal cost and weight update can be achieved for any feed-forward network using HJB equation in a simple yet elegant manner. The proposed approach has been compared with some of the existing best performing learning algorithms. It is found as expected that the proposed approach is faster in convergence in terms of computational time. Some of the benchmark test data such as 8-bit parity, breast cancer and credit approval, as well as 2D Gabor function have been used to validate our claims. The paper also discusses issues related to global optimization. The limitations of popular deterministic weight update laws are critiqued and the possibility of global optimization using HJB formulation is discussed. It is hoped that the proposed algorithm will bring in a lot of interest in researchers working in developing fast learning algorithms and global optimization.