BenSparX introduces the first Bengali conversational speech dataset for Parkinson’s disease detection and pairs it with a robust, explainable machine learning framework. The framework uses diverse acoustic features, systematic feature selection, and advanced classifiers, achieving 95.67% accuracy, 95.62% F1, and 0.990 AUC. SHAP analysis is employed to explain feature contributions, and the model outperforms state‑of‑the‑art methods on other language datasets.
By Riad Hossain, Muhammad Ashad Kabir, Arat Ibne Golam Mowla, Animesh Chandra Roy, Ranjit Kumar Ghosh
The paper introduces BTS-CAFE, a federated domain generalization framework for respiratory sound classification that addresses stethoscope-induced shortcuts. It combines causality-inspired device-style interventions, counterfactual metadata augmentation, and gradient alignment to reduce style–content entanglement and promote device-invariant decision boundaries. Experiments on ICBHI and SPRSound datasets show a 3.69‑point improvement in out-of-distribution performance over the baseline and outperform conventional data augmentation and federated learning methods.
By Heejoon Koo, Yoon Tae Kim, Miika Toikkanen, June-Woo Kim
arXiv:2610.18934v1 Announce Type: cross
Abstract: Conventional deepfake speech detectors primarily target synthesis artifacts and often degrade under unseen attacks or recording conditions. Human spe...
By Vojt\v{e}ch Stan\v{e}k, Michal Fri\v{c}, Kamil Malinka, Anton Firc
arXiv:2607. 00986v1 Announce Type: new Abstract: Automatically detecting stress in speech provides an unobtrusive way to gain insights relevant to behavioral research or clinical assessment.
By Hanna Drimalla, Wieland R. Cremer, Christine Kraus, Oliver T. Wolf
BreathGRU is a semi‑supervised Bidirectional Gated Recurrent Unit framework designed to segment speech and breath events in respiratory audio. It combines acoustic feature extraction, bidirectional recurrent modeling, pseudo‑label refinement, and duration‑constrained Segmental Viterbi decoding to produce accurate speech‑breath segmentation. In evaluations against existing methods, BreathGRU achieved the highest breath event recall, lowest onset‑localisation error, and highest Mean Match Intersection over Union, outperforming large pretrained VAD models such as Silero.
By Sania Fatima Sayed, John W. Holloway, Reyer Zwiggelaar, Faisal I. Rezwan
The paper introduces an interpretable framework for singing emotion recognition that uses a compact set of singing-specific descriptors covering prosody, spectral features, phonetics, and vocal technique. By applying SHAP and GRAD-CAM, the authors link model predictions to measurable acoustic phenomena, showing that vocal technique descriptors reduce cross-singer variability and provide interpretable insights. Out‑of‑domain tests on VocalSet reveal that vibrato and pharyngeal techniques are key emotional cues, with vibrato’s importance depending on articulation rate.
By Fotini Kiriakopoulou, Vassilis Lyberatos, Spyridon Kantarelis, Athanasios Voulodimos, Giorgos Stamou
The paper proposes a linguistically structured multi‑task learning framework for recognizing non‑canonical phonemes by decomposing phoneme prediction into articulatory feature dimensions such as manner, place, and voicing. A hierarchical architecture with task‑specific heads and a cross‑attention fusion module is combined with semi‑supervised Momentum Pseudo‑Labeling and a cascaded training strategy that gradually introduces articulatory tasks. Experiments on the L2‑ARCTIC dataset demonstrate significant improvements over baseline models and produce interpretable error patterns aligned with phonological feature structure.
By Sophia Riaz, Haoze Zheng, Amos Roche, Miyu Zhang, Anamika Ragu, Salvatore Penachio, Kaustav Mukherjee, Aneesh Jonelagadda
The study introduces an automated system for estimating the Verbal Fluency Index (VFI) in individuals with Motor Neuron Disease (MND) by combining ASR (WhisperX) and VAD (Silero) with precise timestamping. Using a unique MND dataset, the approach outperformed traditional acoustic and self‑supervised embedding methods, achieving high predictive accuracy (R² up to 0.9 for P‑words and 0.8 for S‑words). Clinically inspired features were consistently superior, demonstrating the feasibility of automated VFI estimation for monitoring cognitive impairment in MND.
By Bahman Mirheidari, Leslie Ing, Daniel Blackburn, Sharon Abrahams, Christopher McDermott, Heidi Christensen
The study explores acoustic biomarkers derived from spontaneous speech to differentiate Alzheimer's disease patients from healthy controls. Using a structured picture‑description task, low‑level prosodic, spectral, and temporal features were extracted and fed into various machine learning classifiers. The best models achieved clinically meaningful separation between diagnostic groups, indicating that non‑invasive audio screening could aid early detection of dementia.
By Anh L Tran, Arjee Javellana Restar, Bereket Kindo
arXiv:2512. 10120v2 Announce Type: replace-cross Abstract: General-purpose audio representations aim to map acoustically variable instances of the same event to nearby points, resolving content identity in a zero-shot setting.
By Maris Basha, Anja Zai, Sabine Stoll, Richard Hahnloser
arXiv:2607. 23412v1 Announce Type: new Abstract: Electrocardiograms (ECGs) are widely used for cardiovascular risk prediction, yet models often fail to transfer across hospitals because of protocol, population, and measurement differences.
By Jie Lin, Weijie Sun, Sunil V. Kalmady, Anita Khalafbeigi, Abram Hindle, Padma Kaul, Russell Greiner
The paper investigates input representations for multi‑pitch estimation (MPE) in vocal ensembles, comparing the commonly used harmonic constant‑Q transform (HCQT) with a linear short‑time Fourier transform (STFT). Despite HCQT’s frequency‑adaptive resolution, the linear STFT—though fixed in frequency resolution—outperforms HCQT and reduces feature‑extraction cost. The study also finds that longer analysis windows or broader spectral coverage do not improve performance, and that limiting input to the predicted pitch range diminishes the advantage of the linear STFT.
By Junyoung Koh, Hao-Wen Dong