arXiv subjects

Svantje Voit

Publications and source records attributed to Svantje Voit.

4 recordsLinked to original sources

arXiv2026

A Study on Online Mask-based Beamforming Using Per-channel Masking for Spatially Distributed Microphones

Mask-based beamforming is a popular geometry-agnostic approach for speech enhancement, typically applying a single mask across all microphones to estimate the required covariance matrices. While effective for compact arrays, this strategy may be suboptimal for spatially distributed microphones, where signal characteristics may vary strongly across microphones. To effectively capture the spatial diversity across microphones, we extend the mask-based beamformer to a multi-channel formulation, where each microphone is pre-filtered by a separate mask before covariance estimation. To address time-varying acoustic scenes, caused by spectro-temporal nonstationarity, we adopt a frame-causal online implementation with a sliding window. Experiments with simulated compact arrays and distributed microphones show that multi-channel masking yields a benefit over using a single mask when microphone signals differ substantially, while retaining similar performance in compact arrays. We further demonstrate the robustness of the multi-channel masking approach by comparing oracle ideal ratio masks to blind DNN-based mask estimation.

eess.AS↗

arXiv2026

In-the-Loop Training of Deep Feedback Cancellation for Hearing Aids

Acoustic feedback limits the maximum gain in hearing aids. In addition to several approaches based on adaptive filtering, recently a deep-neural-network-based feedback cancellation (DFC) approach has been proposed, which is trained via an open-loop framework. Since open-loop-trained DFC (DFC-OL) can become unstable during inference at high gains, in this paper we propose an in-the-loop-trained DFC (DFC-IL) that integrates the DFC directly into the optimisation loop. This allows the model to be exposed to unstable conditions during training. A two-stage training strategy involving pre-training on stable systems and fine-tuning on a wider gain range enables DFC-IL to learn robust howling reduction. Experimental results on measured feedback paths demonstrate that in scenarios with small gains, the proposed DFC-IL performs similarly to DFC-OL, and both exceed the performance of adaptive filters. In scenarios with high amplification gains, DFC-IL clearly outperforms DFC-OL by maintaining system stability.

eess.AS↗

arXiv2024

Multiplant Nonlinear System Identification by Block-Structured Multikernel Neural Networks in Applications of Interference Cancellation

Problems of linear system identification have closed-form solutions, e.g., using least-squares or maximum-likelihood methods on input-output data. However, already the seemingly simplest problems of nonlinear system identification present more difficulties related to the optimisation of the furrowed error surface. Those cases include the Hammerstein plant with typically a bilinear model representation based on polynomial or Fourier expansion of its nonlinear element. Wiener plants induce actual nonlinearity in the parameters, which further complicates the optimisation. Neural network models and related optimisers are, however, well-prepared to represent and solve nonlinear problems. Unfortunately, the available data for nonlinear system identification might be too diverse to support accurate and consistent model representation. This diversity may refer to different impulse responses and nonlinear functions that arise in different measurements of (different) plants. We therefore propose multikernel neural network models to represent nonlinear plants with a subset of trainable weights shared between different measurements and another subset of plant-specific (i.e., multikernel) weights to adhere to the characteristics of specific measurements. We demonstrate that in this way we can fit neural network models to the diverse data which cannot be done with some standard methods of nonlinear system identification. For model testing, the subset of shared weights of the entire trained model is reused to support the identification and representation of unseen plant measurements, while the plant-specific model weights are readjusted to specifically meet the test data.

eess.SP↗

arXiv2024

On Neural-Network Representation of Wireless Self-Interference for Inband Full-Duplex Communications

Neural network modeling is a key technology of science and research and a platform for deployment of algorithms to systems. In wireless communications, system modeling plays a pivotal role for interference cancellation with specifically high requirements of accuracy regarding the elimination of self-interference in full-duplex relays. This paper hence investigates the potential of identification and representation of the self-interference channel by neural network architectures. The approach is promising for its ability to cope with nonlinear representations, but the variability of channel characteristics is a first obstacle in straightforward application of data-driven neural networks. We therefore propose architectures with a touch of "adaptivity" to accomplish a successful training. For reproducibility of results and further investigations with possibly stronger models and enhanced performance, we document and share our data.

eess.SP↗