Teaching Speech Models to Hear Chaos
A readable breakdown of CIS-BWE, the ACL 2026 paper on chaos-informed speech bandwidth extension.
Read the explainerArticles
Research Articles
Selected papers and research writing on speech reconstruction, bandwidth extension, cyber-physical sensing, low-resource NLP, and applied generative AI.
Featured
Audio reconstruction and CPS security
Nonlinear Framework for Speech Bandwidth Extension
A framework for reconstructing higher-bandwidth speech from limited-bandwidth audio, connecting signal restoration with practical speech enhancement.
Read on arXivHVAC-EAR: Eavesdropping Human Speech Using HVAC Systems
A study of how HVAC-related sensing channels can expose speech information, highlighting security implications for cyber-physical environments.
Reconstructing Intelligible Speech from the Pressure Sensor Data in HVACs
Research on intelligible speech recovery from pressure-sensor data, combining sensor security, physical side channels, and deep learning based reconstruction.
Read on arXivCIS-BWE: Chaos-Informed Speech Bandwidth Extension
A chaos-informed approach to bandwidth extension for speech restoration, accepted to ACL 2026.
Read on arXivGMU Features
Newsletter and news coverage
My work and related lab research have been featured in George Mason University news stories.
Chaos Theory Finds Its Voice
GMU featured the chaos-informed speech reconstruction work behind CIS-BWE and its acceptance to ACL 2026.
Read the GMU storyStaying Ahead of the Current
GMU highlighted related generative AI and signal reconstruction research from the System Design and Security Lab.
Read the GMU storyMore Work
Language, healthcare, and hearables
Milestones in Bengali Sentiment Analysis Leveraging Transformer Models
Survey and analysis of transformer-based Bengali sentiment analysis, including fundamentals, challenges, and future directions.
Read on arXivA High-Fidelity Speech Super Resolution Network
Speech super-resolution using a complex global attention module with spectro-temporal loss.
Read on arXivPrompt Engineering for Hate Speech Detection in Low-Resource Languages
Exploration of prompt engineering strategies for hate speech detection where labeled resources are limited.
Read on arXivA Practical Approach to Power Saving in Hearables
Sub-Nyquist sampling with bandwidth extension as a route toward lower-power hearable audio systems.
Read on arXivEnhancing Diagnostic Accuracy in Bangla
LLM approaches to gender-based medical specialty prediction in Bangla, published in ICCIT 2024 by IEEE.
View on IEEE Xplore