Mounya Elhilali
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Research Interests
Mounya Elhilali has a total of 233 co-authors. Here are the top co-authors by number of shared publications:
Shihab Shamma
Institute for Systems Research, University of Maryland
15 shared publicationsShihab A. Shamma
The Institute for Systems Research, University of Maryland, College Park, Maryland, United States of America; Department of Electrical and Computer Engineering, University of Maryland, College Park, Maryland, United States of America; Département d'Etudes Cognitives, Ecole normale supérieure, Paris, France
12 shared publications- Joel S. Snyder
University of Nevada Las Vegas
11 shared publications Dimitra Emmanouilidou
Department of Electrical and Computer Engineering, Johns Hopkins University, Baltimore, MD, USA
9 shared publications- Jonathan B. Fritz
Center for Neural Science, New York University
9 shared publications - Nathan C. Higgins
Department of Communication Sciences & Disorders, College of Behavioral & Community Sciences, University of South Florida, Tampa
8 shared publications
Latest Publications
- Open Access
- 2 November 2026
- Published by Springer Nature in npj Aging
Review of artificial intelligence technologies for older adult health and wellbeing in NIA AITC Program
- 2 November 2026
- Published by Springer Nature in npj Aging
Open AccessReview of artificial intelligence technologies for older adult health and wellbeing in NIA AITC Program
- Open Access
- 16 October 2026
- Published by Elsevier in NeuroImage
- Vol. 342, 122285
Neural dynamics of auditory attention: Interactions between top-down and bottom-up processes
- 16 October 2026
- Published by Elsevier in NeuroImage
- Vol. 342, 122285
Open AccessNeural dynamics of auditory attention: Interactions between top-down and bottom-up processes
- Open Access
- 1 July 2026
- Published by Acoustical Society of America (ASA) in The Journal of the Acoustical Society of America
- Vol. 160 (1) , 160-173
BioNet-A: Ultrasonic echo representation network for target discrimination using active SONAR
Ultrasonic echoes provide critical cues for object perception, yet their millisecond duration and frequency-specific structure violate assumptions of conventional spectrogram-based convolutional models. Existing biomimetic front-end systems, including auditory spectrograms, cortical wavelets, and the biomimetic BioNet, either under-utilize model capacity on brief echoes or impose frequency shift-invariance suited to vocalizations but detrimental for echo discrimination. This study introduces BioNet5-A, a biomimetic encoder optimized for ultrasonic echoes. BioNet5-A is derived from an autoencoder pretrained on bat vocalizations and incorporates three architectural innovations: (1) spectrotemporal attention to concentrate capacity on the ∼2 ms echo segment and relax frequency-axis invariance; (2) multi-sized convolution/transposed-convolution modules that capture echo structure across multiple scales; and (3) a symmetric, weight-tied encoder–decoder to stabilize training and regularize the biomimetic code. Using a controlled ultrasonic dataset spanning multiple objects and noise conditions, BioNet5-A consistently outperforms auditory spectrogram, cortical wavelet, and the conventional BioNet, and shows improved clustering, discrimination, and robustness. Additionally, representations by the model remain compact and interpretable, aligning with bat midbrain physiology. These results position BioNet5-A as a practical front end for biosonar-inspired sensing and ultrasonic applications.
- 1 July 2026
- Published by Acoustical Society of America (ASA) in The Journal of the Acoustical Society of America
- Vol. 160 (1) , 160-173
Open AccessBioNet-A: Ultrasonic echo representation network for target discrimination using active SONAR
Ultrasonic echoes provide critical cues for object perception, yet their millisecond duration and frequency-specific structure violate assumptions of conventional spectrogram-based convolutional models. Existing biomimetic front-end systems, including auditory spectrograms, cortical wavelets, and the biomimetic BioNet, either under-utilize model capacity on brief echoes or impose frequency shift-invariance suited to vocalizations but detrimental for echo discrimination. This study introduces BioNet5-A, a biomimetic encoder optimized for ultrasonic echoes. BioNet5-A is derived from an autoencoder pretrained on bat vocalizations and incorporates three architectural innovations: (1) spectrotemporal attention to concentrate capacity on the ∼2 ms echo segment and relax frequency-axis invariance; (2) multi-sized convolution/transposed-convolution modules that capture echo structure across multiple scales; and (3) a symmetric, weight-tied encoder–decoder to stabilize training and regularize the biomimetic code. Using a controlled ultrasonic dataset spanning multiple objects and noise conditions, BioNet5-A consistently outperforms auditory spectrogram, cortical wavelet, and the conventional BioNet, and shows improved clustering, discrimination, and robustness. Additionally, representations by the model remain compact and interpretable, aligning with bat midbrain physiology. These results position BioNet5-A as a practical front end for biosonar-inspired sensing and ultrasonic applications.
- Open Access
- 24 May 2026
- Published by Springer Nature in Communications Biology
- Vol. 9 (1)
The shape of attention reflects flexible filtering of natural speech modulations
Perception requires more than passive sensing; it involves prioritizing features relevant to cognitive goals, guided by selective attention. Natural stimuli, such as speech, involve spectrally entangled features, posing demands on attentional mechanisms. A central question is whether attention amplifies all features of the selected input or instead reshapes how stimulus information is represented according to the listener's goal. Here, we recorded EEG while participants performed comprehension or identification tasks on identical continuous speech in noise. To characterize speech cortical representations, we introduce the Modulation Response Function (MRF), a neural analogue of the acoustic modulation spectrum that characterizes cortical sensitivity to spectrotemporal modulations. MRFs revealed distinct relative attentional strategies: comprehension enhanced formant modulations supporting intelligibility, whereas identification emphasized pitch cues linked to voice presence. These differences remained stable across time and listening difficulty, consistent with flexible, goal-dependent attentional filtering. These findings refine future models of how cognitive goals shape speech representations.
- 24 May 2026
- Published by Springer Nature in Communications Biology
- Vol. 9 (1)
Open AccessThe shape of attention reflects flexible filtering of natural speech modulations
Perception requires more than passive sensing; it involves prioritizing features relevant to cognitive goals, guided by selective attention. Natural stimuli, such as speech, involve spectrally entangled features, posing demands on attentional mechanisms. A central question is whether attention amplifies all features of the selected input or instead reshapes how stimulus information is represented according to the listener's goal. Here, we recorded EEG while participants performed comprehension or identification tasks on identical continuous speech in noise. To characterize speech cortical representations, we introduce the Modulation Response Function (MRF), a neural analogue of the acoustic modulation spectrum that characterizes cortical sensitivity to spectrotemporal modulations. MRFs revealed distinct relative attentional strategies: comprehension enhanced formant modulations supporting intelligibility, whereas identification emphasized pitch cues linked to voice presence. These differences remained stable across time and listening difficulty, consistent with flexible, goal-dependent attentional filtering. These findings refine future models of how cognitive goals shape speech representations.
- Open Access
- 12 May 2026
- Published by Center for Open Science in PsyArXiv
Object and Setting Identification in Natural Auditory Scenes
We encounter situations each day that require our auditory system to quickly interpret our surroundings. Auditory scene perception involves complex processes that allow us to identify both the setting and the objects within a scene, which are essential for decision-making, and situational awareness. While there is substantial evidence for distinct cortical regions and pathways supporting visual scene and object recognition, far less is known about how the brain processes complex auditory scenes and objects. This study aimed to determine whether distinct mechanisms underlie auditory setting and object identification and whether these mechanisms interact to aid perception. Participants listened to 200 natural auditory scenes of varying durations (1, 2, and 4 sec) and identified the setting (e.g., café) as well as the objects (e.g., talking, espresso machine, music) within each scene. Overall, performance was highest on the object identification task, and there was a significant interaction between task and scene duration, with a greater benefit of longer durations for the object identification task. Different low- and mid-level acoustic features of the scenes predicted performance on the two tasks. These results suggest that the auditory system employs distinct (and potentially interactive) computations for setting and object identification, allowing for quick interpretation of complex, real-world auditory scenes. The interaction between task performance and scene duration further suggests object and setting identification may operate on different temporal scales; objects might require more time for accurate individuation, while settings may be recognized more efficiently through global properties of scenes, such as openness or naturalness.
- 12 May 2026
- Published by Center for Open Science in PsyArXiv
Open AccessObject and Setting Identification in Natural Auditory Scenes
We encounter situations each day that require our auditory system to quickly interpret our surroundings. Auditory scene perception involves complex processes that allow us to identify both the setting and the objects within a scene, which are essential for decision-making, and situational awareness. While there is substantial evidence for distinct cortical regions and pathways supporting visual scene and object recognition, far less is known about how the brain processes complex auditory scenes and objects. This study aimed to determine whether distinct mechanisms underlie auditory setting and object identification and whether these mechanisms interact to aid perception. Participants listened to 200 natural auditory scenes of varying durations (1, 2, and 4 sec) and identified the setting (e.g., café) as well as the objects (e.g., talking, espresso machine, music) within each scene. Overall, performance was highest on the object identification task, and there was a significant interaction between task and scene duration, with a greater benefit of longer durations for the object identification task. Different low- and mid-level acoustic features of the scenes predicted performance on the two tasks. These results suggest that the auditory system employs distinct (and potentially interactive) computations for setting and object identification, allowing for quick interpretation of complex, real-world auditory scenes. The interaction between task performance and scene duration further suggests object and setting identification may operate on different temporal scales; objects might require more time for accurate individuation, while settings may be recognized more efficiently through global properties of scenes, such as openness or naturalness.
- 1 May 2026
- Published by Acoustical Society of America (ASA) in The Journal of the Acoustical Society of America
- Vol. 159 (5) , 4722-4735
Object and setting identification in natural auditory scenes
Auditory scene perception allows listeners to identify both the setting and the objects within a scene, supporting decision-making and situational awareness. While visual scene and object recognition are well-studied, less is known about how listeners identify settings (e.g., forest) and objects (e.g., birds) in complex auditory environments. This study examined how scene duration influences listeners' ability to identify settings (e.g., café) and objects (e.g., music, talking, espresso machines) in natural auditory scenes. Participants listened to scenes of varying durations (1, 2, and 4 s) and reported the setting and objects present in each scene. Object identification was more accurate than setting identification across durations, but performance on both tasks benefited from increased durations. Several low- and mid-level acoustic features significantly predicted performance, although these models explained relatively little variance overall. These findings suggest that while acoustic structure contributes to performance, differences between object and setting identification likely reflect interacting perceptual and cognitive processes. Setting identification may depend more on integrating information across multiple sound sources and global scene properties (e.g., openness or naturalness), whereas object identification may rely more on segregation and recognition of individual sound sources and their features.
- 1 May 2026
- Published by Acoustical Society of America (ASA) in The Journal of the Acoustical Society of America
- Vol. 159 (5) , 4722-4735
Object and setting identification in natural auditory scenes
Auditory scene perception allows listeners to identify both the setting and the objects within a scene, supporting decision-making and situational awareness. While visual scene and object recognition are well-studied, less is known about how listeners identify settings (e.g., forest) and objects (e.g., birds) in complex auditory environments. This study examined how scene duration influences listeners' ability to identify settings (e.g., café) and objects (e.g., music, talking, espresso machines) in natural auditory scenes. Participants listened to scenes of varying durations (1, 2, and 4 s) and reported the setting and objects present in each scene. Object identification was more accurate than setting identification across durations, but performance on both tasks benefited from increased durations. Several low- and mid-level acoustic features significantly predicted performance, although these models explained relatively little variance overall. These findings suggest that while acoustic structure contributes to performance, differences between object and setting identification likely reflect interacting perceptual and cognitive processes. Setting identification may depend more on integrating information across multiple sound sources and global scene properties (e.g., openness or naturalness), whereas object identification may rely more on segregation and recognition of individual sound sources and their features.
- 1 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Biomedical Engineering
- Vol. 73 (5) , 1696-1704
Multi-Stage Respiratory Sound Analysis: Confidence-Driven Wheeze and Crackle Detection
Objective: Accurate detection of adventitious respiratory sounds, such as wheezes and crackles, is essential for diagnosing and managing respiratory conditions. This study introduces a multi-stage, confidence-driven framework for automated pediatric auscultation analysis, performing a three-way classification of normal, wheeze, and crackle sounds to improve diagnostic accuracy. Methods: We develop a comprehensive pipeline integrating anomaly-specific segment selection, segment-level classification, and confidence-based fusion. Our contrastive variational recurrent neural network (CVRNN) enhances feature extraction, while a confidence-weighted aggregation strategy refines final predictions. The system is validated using a diverse pediatric dataset from 742 subjects (aged 1-59 months) from seven countries. Results: The multi-level framework is evaluated across three stages. The anomaly-specific segment selection achieves 98.47 $\%$ recall, identifying adventitious regions. Next, segment-level classifiers improve sensitivity, achieving balanced accuracies of 72.15 $\%$ (wheeze) and 68.1 $\%$ (crackle). This performance surpasses state of the art systems on the same dataset and demonstrates enhanced balanced performance in detecting both crackle and wheeze sounds, which present different challenges to automated systems given their markedly different acoustic profiles. Finally, the confident-driven fusion outperforms traditional aggregation methods, yielding a final three-way classification of 62.12 $\%$. Conclusion: Our confidence-based multi-stage approach enhances automated respiratory sound classification by prioritizing high-certainty segment predictions, aligning with expert physician annotations. Significance: This framework advances computer-aided respiratory diagnostics, improving early detection and monitoring of pediatric respiratory conditions. By integrating expert-inspired segmentation with machine learning-driven confidence estimation, it has potential to enhance clinical workflows and screening for pulmonary diseases, particularly in resource-limited settings.
- 1 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Biomedical Engineering
- Vol. 73 (5) , 1696-1704
Multi-Stage Respiratory Sound Analysis: Confidence-Driven Wheeze and Crackle Detection
Objective: Accurate detection of adventitious respiratory sounds, such as wheezes and crackles, is essential for diagnosing and managing respiratory conditions. This study introduces a multi-stage, confidence-driven framework for automated pediatric auscultation analysis, performing a three-way classification of normal, wheeze, and crackle sounds to improve diagnostic accuracy. Methods: We develop a comprehensive pipeline integrating anomaly-specific segment selection, segment-level classification, and confidence-based fusion. Our contrastive variational recurrent neural network (CVRNN) enhances feature extraction, while a confidence-weighted aggregation strategy refines final predictions. The system is validated using a diverse pediatric dataset from 742 subjects (aged 1-59 months) from seven countries. Results: The multi-level framework is evaluated across three stages. The anomaly-specific segment selection achieves 98.47 $\%$ recall, identifying adventitious regions. Next, segment-level classifiers improve sensitivity, achieving balanced accuracies of 72.15 $\%$ (wheeze) and 68.1 $\%$ (crackle). This performance surpasses state of the art systems on the same dataset and demonstrates enhanced balanced performance in detecting both crackle and wheeze sounds, which present different challenges to automated systems given their markedly different acoustic profiles. Finally, the confident-driven fusion outperforms traditional aggregation methods, yielding a final three-way classification of 62.12 $\%$. Conclusion: Our confidence-based multi-stage approach enhances automated respiratory sound classification by prioritizing high-certainty segment predictions, aligning with expert physician annotations. Significance: This framework advances computer-aided respiratory diagnostics, improving early detection and monitoring of pediatric respiratory conditions. By integrating expert-inspired segmentation with machine learning-driven confidence estimation, it has potential to enhance clinical workflows and screening for pulmonary diseases, particularly in resource-limited settings.
- Open Access
- 1 April 2026
- Published by Acoustical Society of America (ASA) in JASA Express Letters
- Vol. 6 (4)
Auditory deviance detection across time scales: Effects of local and global context
Auditory deviance detection reflects the ability to identify violations of statistical regularities in sound sequences and is influenced by stimulus properties and contextual expectations. Using a priming paradigm in which listeners were exposed to different pitch distributions prior to deviant detection trials, we examined how performance varied across the experiment (global context) and relative to short-term priors (local context). Results show that listeners accumulate statistical information across distinct contextual time scales, where local and global context influence deviance detection. Computational modeling using a predictive coding framework is consistent with these effects, highlighting the importance of integrating contextual information across scales in dynamic acoustic environments.
- 1 April 2026
- Published by Acoustical Society of America (ASA) in JASA Express Letters
- Vol. 6 (4)
Open AccessAuditory deviance detection across time scales: Effects of local and global context
Auditory deviance detection reflects the ability to identify violations of statistical regularities in sound sequences and is influenced by stimulus properties and contextual expectations. Using a priming paradigm in which listeners were exposed to different pitch distributions prior to deviant detection trials, we examined how performance varied across the experiment (global context) and relative to short-term priors (local context). Results show that listeners accumulate statistical information across distinct contextual time scales, where local and global context influence deviance detection. Computational modeling using a predictive coding framework is consistent with these effects, highlighting the importance of integrating contextual information across scales in dynamic acoustic environments.
- Open Access
- 5 February 2026
- Published by Elsevier in NeuroImage
- Vol. 328, 121785
Perception of dynamic multi-speaker auditory scenes under different modes of attention
- 5 February 2026
- Published by Elsevier in NeuroImage
- Vol. 328, 121785
Open AccessPerception of dynamic multi-speaker auditory scenes under different modes of attention