Israel Cohen

Modern hands-free and wearable communication devices employ multiple microphones and loudspeakers, leading to the joint presence of acoustic echo, background noise, and desired speech signals. While acoustic echo cancellation (AEC) and beamforming are commonly combined to address this challenge, existing architectures face a trade-off between computational complexity, stability, and adaptability. In particular, adaptive beamforming approaches require repeated estimation and inversion of covariance matrices, incurring high computational cost and introducing potential sensitivity to time-varying conditions. Conversely, fixed beamformers reduce online complexity and improve stability, but their performance can degrade when the acoustic scene differs from the calibration condition. In this work, we investigate low-complexity AEC–beamforming architectures that combine fixed minimum-variance distortionless response (MVDR) beamforming with adaptive AEC. Since the ordering of these stages yields two inequivalent architectures, we evaluate two configurations: AEC followed by beamforming (AEC-BF) and beamforming followed by AEC (BF-AEC). To reduce dependence on a single head pose in wearable devices, we use an offline orientation-averaged calibration strategy in which the undesired-signal covariance matrix and, when required, the relative echo transfer functions (RETFs) are estimated from calibration measurements averaged across multiple head orientations. The proposed methods are evaluated using real-device recordings from a six-microphone wearable device. The results show a clear architectural asymmetry: the fixed BF-AEC configuration achieves the highest average echo return loss enhancement (ERLE) and perceptual evaluation of speech quality (PESQ), with substantially lower online complexity than the fully adaptive baseline, whereas the fixed AEC-BF configuration provides a higher signal-to-distortion ratio (SDR) in the evaluated experiment. Additional calibration experiments show that orientation-averaged RETF calibration provides partial generalization across the measured head orientations, but also that the RETFs are not fully orientation-invariant. Overall, the results indicate that fixed BF-AEC provides a favorable trade-off between echo suppression, stability, and online complexity under the evaluated real-recording conditions.

Speech separation with incremental learning (SSIL), which facilitates the continuous adaptation of speech separation models to new languages, remains a critical yet underexplored area. In this paper, we propose a novel framework, termed discrete Fourier transform Adapters for Speech Separation with incremental learning (FASS), designed to mitigate catastrophic forgetting by training orthogonal Fourier-domain adapters using a parameter expansion-fusion strategy. This approach stems from our analysis of how acquiring new tasks interferes with retaining prior knowledge. Specifically, FASS converts sparse Fourier-domain parameters into dense parameter-domain weights via the inverse discrete Fourier transform (IDFT), which could preserve previous knowledge without necessitating task identifiers or additional storage. Furthermore, an extended variant, FASS-random (FASS-r), enhances scalability by randomly assigning indices to the learnable frequencies while maintaining theoretical performance. Comprehensive theoretical analyses and extensive experimental results substantiate the effectiveness of our approach.

Sensing audio using non-acoustic modalities such as millimeter-wave radar and laser-based systems has emerged as an active research area with significant implications for privacy, security, and robust speech processing. These approaches recover speech-related information from vibration measurements captured by non-acoustic sensing modalities. Prior work spans a wide range of techniques, from classical signal-processing pipelines to modern machine-learning and deep-learning models, enabling applications such as speech reconstruction, eavesdropping, automatic speech recognition, and noise-robust enhancement. Some systems rely on radar or laser sensing as a standalone audio surrogate, while others fuse radar-derived features with microphone signals to improve robustness in noisy or non-line-of-sight environments. Experimental results across the literature demonstrate that recovering intelligible speech or discriminative speech features from radar or laser-sensed vibrations is feasible under controlled conditions. However, performance remains sensitive to practical factors including sensing distance, object material and geometries, environmental interference, multipath effects, and task complexity. Not all speech-related tasks are reliably solved, particularly in unconstrained real-world scenarios. Overall, the field is rapidly evolving, with open challenges in robustness, generalization, and deployment, offering several promising directions for future research.

Region-of-interest (ROI) beamforming offers a promising solution for enhancing speech perception in noisy environments using assistive hearing devices, particularly those equipped with microphone arrays mounted on glasses. However, when paired with open-ear air-conduction loudspeakers, these systems encounter significant challenges with acoustic feedback and amplification of the user's own voice. This paper addresses these challenges by introducing an adaptive broadband time-domain ROI beamforming approach. We derive a ROI distortion constraint to direct the beamformer toward a target speaker located within the ROI while minimizing distortion. We propose the regularized least-distortion minimum-power (RLDMP) ROI beamformer as an optimal convolutional filter that minimizes the variance of the beamformer output subject to the ROI distortion constraint. As a baseline, we also formulate maximum-gain-based ROI beamformers under various constraints for ROI distortion and interference suppression. To address time-variant acoustic feedback and adapt to changes in room acoustics, we introduce an adaptive feedback suppression solution that employs a smoothing mechanism on estimates of the sensor-sample correlation matrix. To prevent own-voice periods from contaminating the adaptive process, a spatial-dominance-controlled strategy is proposed, along with a gain-control logic to manage own-voice amplification. Experimental evaluations using a simulated hearing glasses system demonstrate that the proposed approach successfully increases the maximum stable gain (MSG) while maintaining high-quality speech reinforcement. We also discuss the system's physical limitations and propose strategies to regain stability.

One of the key challenges in array signal processing is the lack of precise source location information, which can arise from estimation errors or the presence of a moving source. Conventional beamformers, which rely on accurate spatial information, perform suboptimally under such conditions. Region-of-interest (ROI) beamformers address this limitation by focusing on a spatial region rather than a single direction. In this work, we introduce a new ROI beamformer based on the generalized sidelobe canceller (GSC) framework with time-domain convolutive filters. The proposed beamformer can concentrate on a specified ROI while suppressing noise and interference from other directions. To evaluate the efficacy of the proposed beamformer, we derive several performance metrics. Then we analyze various blocking matrix designs, focusing on signal leakage as a key factor in interference suppression without distorting the desired signal. Furthermore, an adaptive implementation based on the recursive least-squares (RLS) algorithm is developed, along with a low-complexity variant using the nearest Kronecker product (NKP) approximation. Simulation results demonstrate the effectiveness of the proposed ROI-GSC beamformer.

The proliferation of satellite constellations, coupled with reduced tasking latency and diverse sensor capabilities, has expanded the opportunities for automated Earth observation. This paper introduces a fully automated Tip-and-Cue framework designed for satellite imaging tasking and scheduling. In this context, tips are generated from external data sources or analyses of prior satellite imagery, identifying spatiotemporal targets and prioritizing them for downstream planning. Corresponding cues are the imaging tasks formulated in response, which incorporate sensor constraints, timing requirements, and utility functions. The system autonomously generates candidate tasks, optimizes their scheduling across multiple satellites using continuous utility functions that reflect the expected value of each observation, and processes the resulting imagery using artificial-intelligence-based models, including object detectors and vision-language models. Structured visual reports are generated to support both interpretability and the identification of new insights for downstream tasking. The efficacy of the framework is demonstrated through a maritime vessel tracking scenario, utilizing Automatic Identification System (AIS) data for trajectory prediction, targeted observations, and the generation of actionable outputs. Maritime vessel tracking is a widely researched application, often used to benchmark novel approaches to satellite tasking, forecasting, and analysis. The system is extensible to broader applications such as smart-city monitoring and disaster response, where timely tasking and automated analysis are critical.