# Neue Publikationen in Ziel-Venues – 2026-11-16 Neue Werke: 23 · davon mit DE-Affiliation: 23 Venues: 60 · abgefragt ab publication_date 2026-10-17 · letzter Lauf: 2026-11-16T18:09:50 ## IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) — 22 neu · _conference_ ### Kalman Filter Based Linear Deformable for Retinal Vessel Segmentation ⚑ DE - **DOI:** [10.1109/icassp55912.2026.11462647](https://doi.org/10.1109/icassp55912.2026.11462647) - **Metadaten:** Erschienen: 2026-04-21 · S. 5891-5895 · OpenAlex seit 2026-04-22 - **DE-Institutionen:** TUM Klinikum; Technical University of Munich - **Autoren:** - Zhihao Zhao — TUM Klinikum (DE); Technical University of Munich (DE) - Yinzheng Zhao — TUM Klinikum (DE); Technical University of Munich (DE) - Junjie Yang — TUM Klinikum (DE); Technical University of Munich (DE) - Quanmin Liang — Sun Yat-sen University (CN) - Daniel Zapp — TUM Klinikum (DE); Technical University of Munich (DE) - Kai Huang — Sun Yat-sen University (CN) - Mohammad Ali Nasseri — TUM Klinikum (DE); Technical University of Munich (DE) - **Topics:** Retinal Imaging and Analysis; Medical Image Segmentation Techniques; Advanced Neural Network Applications - **Keywords:** retinal vessel segmentation; Kalman filter; cross-attention; U-Net; topological continuity; fundus images; optical coherence tomography angiography - **Abstract:** In the realm of ophthalmic imaging, accurate vascular segmentation is paramount for diagnosing and managing various eye diseases. Contemporary deep learning-based vascular segmentation models rival human accuracy but still face substantial challenges in accurately segmenting minuscule blood vessels in neural network applications. In this paper, we propose a novel network (KaLDeX) for vascular segmentation leveraging a Kalman filter based linear deformable cross attention (LDCA) module integrated within a UNet++ framework. Our approach is based on two key components: Kalman filter (KF) based linear deformable convolution (LD) and cross-attention (CA) modules. The LD module is designed to adaptively adjust the focus on thin vessels that might be overlooked in standard convolution. The CA module improves the global understanding of vascular structures by aggregating the detailed features from the LD module with the high-level features from the UNet++ architecture. Finally, we adopt a topological loss function based on persistent homology to constrain the topological continuity of the segmentation. The proposed method is evaluated on retinal fundus image datasets (DRIVE, CHASE BD1, and STARE) as well as the 3 mm and 6 mm of the OCTA-500 dataset, achieving an average accuracy (ACC) of 97.25%, 97.77%, 97.85%, 98.89%, and 98.21%, respectively. ### Blind Direction-Dependent Acoustic Parameter Estimation Using Smart Glasses ⚑ DE - **DOI:** [10.1109/icassp55912.2026.11460445](https://doi.org/10.1109/icassp55912.2026.11460445) - **Metadaten:** Erschienen: 2026-04-21 · S. 22187-22191 · OpenAlex seit 2026-04-22 - **DE-Institutionen:** Friedrich-Alexander-Universität Erlangen-Nürnberg - **Autoren:** - Philipp Götz — Friedrich-Alexander-Universität Erlangen-Nürnberg (DE) - Sebastià V. Amengual Garí — Meta Reality Labs,Redmond,Washington,USA - Paul Thomas Calamia — Meta Reality Labs,Redmond,Washington,USA - Ishwarya Ananthabhotla — Meta Reality Labs,Redmond,Washington,USA - Andrew F. Francl — Meta Reality Labs,Redmond,Washington,USA - Carl Schissler — Meta Reality Labs,Redmond,Washington,USA - Emanuël A. P. Habets — Friedrich-Alexander-Universität Erlangen-Nürnberg (DE) - **Topics:** Speech and Audio Processing; Music and Audio Processing; Direction-of-Arrival Estimation Techniques - **Keywords:** blind source separation; acoustic parameter estimation; smart glasses; reverberant speech; head rotation - **Abstract:** We address the task of blind, direction-dependent estimation of acoustic parameters from reverberant speech captured with smart glasses equipped with a compact microphone array. Such estimates are valuable for realistic rendering of virtual sound sources in auditory augmented reality. To overcome the limited spatial resolution inherent to compact arrays, the proposed method exploits natural head rotations by aggregating spatial information across multiple viewing orientations. The approach is validated by estimating the direction-dependent decay time T20and directional acoustic energy. ### Loose Coupling of Spectral and Spatial Models for Multi-Channel Diarization and Enhancement of Meetings in Dynamic Environments ⚑ DE - **DOI:** [10.1109/icassp55912.2026.11463540](https://doi.org/10.1109/icassp55912.2026.11463540) - **Metadaten:** Erschienen: 2026-04-21 · S. 16852-16856 · OpenAlex seit 2026-04-22 - **DE-Institutionen:** Paderborn University - **Autoren:** - Adrian Meise — Paderborn University (DE) - Tobias Cord-Landwehr — Paderborn University (DE) - Christoph Boeddeker — Paderborn University (DE) - Marc Delcroix — NTT (Japan) (JP) - Tomohiro Nakatani — NTT (Japan) (JP) - Reinhold Haeb‐Umbach — Paderborn University (DE) - **Topics:** Speech and Audio Processing; Speech Recognition and Synthesis; Music and Audio Processing - **Keywords:** speaker diarization; multichannel audio; microphone array; meeting transcription; sound enhancement; dynamic environments - **Abstract:** Sound capture by microphone arrays opens the possibility to exploit spatial, in addition to spectral, information for diarization and signal enhancement, two important tasks in meeting transcription. However, there is no one-to-one mapping of positions in space to speakers if speakers move. Here, we address this by proposing a novel joint spatial and spectral mixture model, whose two submodels are loosely coupled by modeling the relationship between speaker and position index probabilistically. Thus, spatial and spectral information can be jointly exploited, while at the same time allowing for speakers speaking from different positions. Experiments on the LibriCSS data set with simulated speaker position changes show great improvements over tightly coupled subsystems. ### Graph Distribution-Valued Signals: A Wasserstein Space Perspective ⚑ DE - **DOI:** [10.1109/icassp55912.2026.11462807](https://doi.org/10.1109/icassp55912.2026.11462807) - **Metadaten:** Erschienen: 2026-04-21 · S. 166-170 · OpenAlex seit 2025-10-19 - **DE-Institutionen:** Technical University of Munich - **Autoren:** - Yanan Zhao — Nanyang Technological University (SG) - Feng Ji — Nanyang Technological University (SG) - Xingchao Jian — Technical University of Munich (DE) - Wee Peng Tay — Nanyang Technological University (SG) - **Topics:** Advanced Graph Neural Networks; Graph Theory and Algorithms; Mathematical Analysis and Transform Methods - **Keywords:** graph signal processing; Wasserstein space; probability distributions; uncertainty modeling; graph filtering; random signals - **Abstract:** We introduce a novel framework for graph signal processing (GSP) that models signals as graph distribution-valued signals (GDSs), which are probability distributions in the Wasserstein space. This approach overcomes key limitations of classical vector-based GSP, including the assumption of synchronous observations over vertices, the inability to capture uncertainty, and the requirement for strict correspondence in graph filtering. By representing signals as distributions, GDSs naturally encode uncertainty and stochasticity, while strictly generalizing traditional graph signals. We establish a systematic dictionary mapping core GSP concepts to their GDS counterparts, demonstrating that classical definitions are recovered as special cases. The effectiveness of the framework is validated through graph filter learning for prediction tasks, supported by experimental results. ### I-DCCRN-VAE: An Improved Deep Representation Learning Framework for Complex VAE-Based Single-Channel Speech Enhancement ⚑ DE - **DOI:** [10.1109/icassp55912.2026.11464594](https://doi.org/10.1109/icassp55912.2026.11464594) - **Metadaten:** Erschienen: 2026-04-21 · S. 15632-15636 · OpenAlex seit 2025-10-17 - **DE-Institutionen:** Carl von Ossietzky Universität Oldenburg; Hearing4all - **Autoren:** - Jiatong Li — Carl von Ossietzky Universität Oldenburg (DE); Hearing4all (DE) - Simon Doclo — Carl von Ossietzky Universität Oldenburg (DE); Hearing4all (DE) - **Topics:** Speech and Audio Processing; Speech Recognition and Synthesis; Image and Signal Denoising Methods - **Keywords:** single-channel speech enhancement; beta-VAE; skip connections; latent space regularization; adversarial training; fine-tuning - **Abstract:** Recently, a complex variational autoencoder (VAE)-based single-channel speech enhancement system based on the DCCRN architecture has been proposed. In this system, a noise suppression VAE (NSVAE) learns to extract clean speech representations from noisy speech using pretrained clean speech and noise VAEs with skip connections. In this paper, we improve DCCRN-VAE by incorporating three key modifications: 1) removing the skip connections in the pretrained VAEs to encourage more informative speech and noise latent representations; 2) using β-VAE in pretraining to better balance reconstruction and latent space regularization; and 3) a NSVAE generating both speech and noise latent representations. Experiments show that the proposed system achieves comparable performance as the DCCRN and DCCRN-VAE baselines on the matched DNS3 dataset but outperforms the baselines on mismatched datasets (WSJ0-QUT, Voicebank-DEMEND), demonstrating improved generalization ability. In addition, an ablation study shows that a similar performance can be achieved with classical fine-tuning instead of adversarial training, resulting in a simpler training pipeline. ### RankNB: Ranking-Aware Direct Preference Optimization for Alignment of a Nanobody Diffusion Model ⚑ DE - **DOI:** [10.1109/icassp55912.2026.11462703](https://doi.org/10.1109/icassp55912.2026.11462703) - **Metadaten:** Erschienen: 2026-04-21 · S. 3246-3250 · OpenAlex seit 2026-04-22 - **DE-Institutionen:** Merck KGaA, Darmstadt (Germany) - **Autoren:** - Yikai Wu — Fudan University (CN) - Di Wu — Shanghai Jiao Tong University (CN) - Jia Shu — Shanghai Jiao Tong University (CN) - Stephanie Maria Linker — Merck KGaA, Darmstadt (Germany) (DE) - Karin Hrovatin — Merck KGaA, Darmstadt (Germany) (DE) - Mathias Winkel — Merck KGaA,AI&Quantum Lab,Darmstadt,Germany - Feng Tan — Merck KGaA,AI&Quantum Lab,Darmstadt,Germany - **Topics:** Protein Structure and Dynamics; Monoclonal and Polyclonal Antibodies Research; Computational Drug Discovery Methods - **Keywords:** nanobodies; diffusion models; preference optimization; protein structure alignment; generative models ### A Novel Iterative OTFS Detector Based on Local L-MMSE and Global Message Passing ⚑ DE - **DOI:** [10.1109/icassp55912.2026.11462455](https://doi.org/10.1109/icassp55912.2026.11462455) - **Metadaten:** Erschienen: 2026-04-21 · S. 21501-21505 · OpenAlex seit 2026-04-22 - **DE-Institutionen:** Technische Universität Berlin - **Autoren:** - Ruohai Yang — Huazhong University of Science and Technology (CN) - Shuangyang Li — Technische Universität Berlin (DE) - Han Yu — Technische Universität Berlin (DE) - Zhiqiang Wei — Xi'an Jiaotong University (CN) - Kai Wan — Huazhong University of Science and Technology (CN) - Giuseppe Caire — Technische Universität Berlin (DE) - **Topics:** PAPR reduction in OFDM; Advanced Wireless Communication Techniques; Optical Network Technologies - **Keywords:** orthogonal time frequency space; iterative detection; message passing; doubly selective channels; high-mobility channels; linear equalization; low-complexity detection - **Abstract:** Orthogonal time frequency space (OTFS) modulation has attracted significant attention due to its robust performance over high-mobility channels. However, conventional detectors, like linear equalizers and message passing (MP), suffer from noise enhancement or fail in complex doubly-selective channels. We propose a hybrid iterative detection framework combining linear minimum mean square error (L-MMSE) estimation with MP. A key novelty is the delay-Doppler commutation precoder (DDCP), which structures the effective channel into sparsely interconnected, locally dense blocks. This enables a hybrid strategy: low-dimensional L-MMSE handles intra-block interference locally, while MP handles sparse inter-block connections globally. The proposed scheme achieves near-optimal error performance with low complexity. ### Do We Need EMA for Diffusion-Based Speech Enhancement? Toward A Magnitude-Preserving Network Architecture ⚑ DE - **DOI:** [10.1109/icassp55912.2026.11462395](https://doi.org/10.1109/icassp55912.2026.11462395) - **Metadaten:** Erschienen: 2026-04-21 · S. 15647-15651 · OpenAlex seit 2025-10-10 - **DE-Institutionen:** Universität Hamburg - **Autoren:** - Julius Richter — Universität Hamburg (DE) - Danilo de Oliveira — Universität Hamburg (DE) - Timo Gerkmann — Universität Hamburg (DE) - **Topics:** Speech and Audio Processing; Speech Recognition and Synthesis; Image and Signal Denoising Methods - **Keywords:** speech enhancement; diffusion models; exponential moving average; Schrödinger bridge; skip connections; environmental noise prediction; VoiceBank-DEMAND - **Abstract:** We study diffusion-based speech enhancement using a Schrödinger bridge formulation and extend the EDM2 framework to this setting. We employ time-dependent preconditioning of network inputs and outputs to stabilize training and explore two skip-connection configurations that allow the network to predict either environmental noise or clean speech. To control activation and weight magnitudes, we adopt a magnitude-preserving architecture and learn the contribution of the noisy input within each network block for improved conditioning. We further analyze the impact of exponential moving average (EMA) parameter smoothing by approximating different EMA profiles post training, finding that, unlike in image generation, short or absent EMA consistently yields better speech enhancement performance. Experiments on VoiceBank-DEMAND and EARS-WHAM demonstrate competitive signal-to-distortion ratios and perceptual scores, with the two skip-connection variants exhibiting complementary strengths. These findings provide new insights into EMA behavior, magnitude preservation, and skip-connection design for diffusion-based speech enhancement. ### Improving the Reasoning of Multi-Image Grounding in MLLMs via Reinforcement Learning ⚑ DE - **DOI:** [10.1109/icassp55912.2026.11461396](https://doi.org/10.1109/icassp55912.2026.11461396) - **Metadaten:** Erschienen: 2026-04-21 · S. 12667-12671 · OpenAlex seit 2025-10-10 - **DE-Institutionen:** Technical University of Munich - **Autoren:** - Bob Zhang — Xiaohongshu (Little Red Book) (China) (CN) - Haoran Li — Xiaohongshu (Little Red Book) (China) (CN) - Tao Zhang — Wuhan University (CN) - Jianan Li — Technical University of Munich (DE) - Cilin Yan — Xiaohongshu (Little Red Book) (China) (CN) - Xikai Liu — Xiaohongshu (Little Red Book) (China) (CN) - Jiayin Cai — Xiaohongshu (Little Red Book) (China) (CN) - Yanbin Hao — Hefei University of Technology (CN) - **Topics:** Multimodal Machine Learning Applications; Advanced Neural Network Applications; Domain Adaptation and Few-Shot Learning - **Keywords:** multimodal large language models; reinforcement learning; post-training; low-rank adaptation; rejection sampling; visual grounding - **Abstract:** Multimodal Large Language Models (MLLMs) perform well in single-image visual grounding but struggle with real-world tasks that demand cross-image reasoning and multi-modal instructions. To address this, we adopt a reinforcement learning (RL) based post-training strategy for MLLMs in multi-image grounding tasks. We first synthesize high-quality chain-of-thought (CoT) data for cold-start initialization, followed by supervised fine-tuning (SFT) using low-rank adaptation (LoRA). Subsequently, we apply rejection sampling with the merged SFT model to curate reliable RL data and use rule-based RL to guide the model toward optimal reasoning paths. Extensive experiments demonstrate the effectiveness of our approach, achieving +9.04% on MIG-Bench and +4.41% on average across seven out-of-domain benchmarks. ### Mixture-of-Experts Based Soft-Label Learning for Multi-Label Speech Emotion Recognition ⚑ DE - **DOI:** [10.1109/icassp55912.2026.11460912](https://doi.org/10.1109/icassp55912.2026.11460912) - **Metadaten:** Erschienen: 2026-04-21 · S. 17172-17176 · OpenAlex seit 2026-04-22 - **DE-Institutionen:** Technical University of Munich - **Autoren:** - Yandi Zheng — Tianjin Normal University (CN) - Xinzhou Xu — Nanjing University of Posts and Telecommunications (CN) - Ziping Zhao — Tianjin Normal University (CN) - Björn W. Schuller — Technical University of Munich (DE) - **Topics:** Emotion and Mood Recognition; Speech Recognition and Synthesis; Speech and Audio Processing - **Keywords:** mixture of experts; majority rule; plurality rule; emotional ambiguity - **Abstract:** Multi-label Speech Emotion Recognition (SER) aims to recognise one or more co-occurring emotions from speech, due to emotional ambiguity and human-perception subjectivity. Existing endeavors usually rely on Majority Rule (MR) and Plurality Rule (PR) to obtain a consensus single label for each sample, which fails to capture the correlation between multiple emotional categories. In this regard, we propose a Mixture-of-Experts (MoE) based soft-label learning for multi-label SER. The proposed approach contains a speech representation extracting module and an MoE module, which captures discriminative information among multiple annotators’ annotations. Experimental results on emotional speech datasets demonstrate that, the proposed approach outperforms compared state-of-the-art ones, showcasing its effectiveness for multi-label SER tasks. ### Adapter-Based Multi-Agent AVSR Extension for Pre-Trained ASR Models ⚑ DE - **DOI:** [10.1109/icassp49660.2025.10889566](https://doi.org/10.1109/icassp49660.2025.10889566) - **Metadaten:** Erschienen: 2025-03-12 · S. 1-5 · OpenAlex seit 2025-03-13 - **DE-Institutionen:** University of Technology Nuremberg - **Autoren:** - Christopher Simic — University of Technology Nuremberg (DE) - Korbinian Riedhammer — University of Technology Nuremberg (DE) - Tobias Bocklet — University of Technology Nuremberg (DE) - **Topics:** Speech and Audio Processing; Speech Recognition and Synthesis; Speech and dialogue systems - **Keywords:** audio-visual speech recognition; Whisper; LoRA adapters; parameter-efficient fine-tuning; audiovisual integration; noise robustness; noise level adaptation - **Abstract:** We present an approach to Audio-Visual Speech Recognition that builds on a pre-trained Whisper model. To infuse visual information into this audio-only model, we extend it with an AV fusion module and LoRa adapters, one of the most up-to-date adapter approaches. One advantage of adapter-based approaches, is that only a relatively small number of parameters are trained, while the basic model remains unchanged. Common AVSR approaches train single models to handle several noise categories and noise levels simultaneously. Taking advantage of the lightweight nature of adapter approaches, we train noise-scenario-specific adapter-sets, each covering individual noise-categories or a specific noise-level range. The most suitable adapter-set is selected by previously classifying the noise-scenario. This enables our models to achieve an optimum coverage across different noise-categories and noise-levels, while training only a minimum number of parameters.Compared to a full fine-tuning approach with SOTA performance our models achieve almost comparable results over the majority of the tested noise-categories and noise-levels, with up to 88.5% less trainable parameters. Our approach can be extended by further noise-specific adapter-sets to cover additional noise scenarios. It is also possible to utilize the underlying powerful ASR model when no visual information is available, as it remains unchanged. ### HRTF Estimation using a Score-based Prior ⚑ DE - **DOI:** [10.1109/icassp49660.2025.10888032](https://doi.org/10.1109/icassp49660.2025.10888032) - **Metadaten:** Erschienen: 2025-03-12 · S. 1-5 · OpenAlex seit 2025-03-13 - **DE-Institutionen:** Universität Hamburg - **Autoren:** - Etienne Thuillier — Aalto University (FI) - Jean-Marie Lemercier — Universität Hamburg (DE) - Eloi Moliner — Aalto University (FI) - Timo Gerkmann — Universität Hamburg (DE) - Vesa Välimäki — Aalto University (FI) - **Topics:** Speech and Audio Processing; Speech Recognition and Synthesis; Music and Audio Processing - **Keywords:** score-based diffusion models; data-driven priors; room impulse response estimation; reverberant environments; human speech; log-likelihood approximation; spatial audio - **Abstract:** We present a head-related transfer function (HRTF) estimation method which relies on a data-driven prior given by a score-based diffusion model. The HRTF is estimated in reverberant environments using natural excitation signals, e.g. human speech. The impulse response of the room is estimated along with the HRTF by optimizing a parametric model of reverberation based on the statistical behaviour of room acoustics. The posterior distribution of HRTF given the reverberant measurement and excitation signal is modelled using the score-based HRTF prior and a log-likelihood approximation. We show that the resulting method outperforms several baselines, including an oracle recommender system that assigns the optimal HRTF in our training set based on the smallest distance to the true HRTF at the given direction of arrival. In particular, we show that the diffusion prior can account for the large variability of high-frequency content in HRTFs. ### Nonlinear Precoding in the RIS-Aided MIMO Broadcast Channel ⚑ DE - **DOI:** [10.1109/icassp49660.2025.10889796](https://doi.org/10.1109/icassp49660.2025.10889796) - **Metadaten:** Erschienen: 2025-03-12 · S. 1-5 · OpenAlex seit 2025-03-13 - **DE-Institutionen:** Technical University of Munich - **Autoren:** - Dominik Semmler — Technical University of Munich (DE) - Michael Joham — Technical University of Munich (DE) - Wolfgang Utschick — Technical University of Munich (DE) - **Topics:** Advanced Wireless Communication Technologies; Millimeter-Wave Propagation and Modeling; Advanced Antenna and Metasurface Technologies - **Keywords:** Tomlinson-Harashima precoding; reconfigurable intelligent surface; MIMO broadcast channel; nonlinear precoding; dirty paper coding; phase shift optimization; discrete phase shifts; statistical channel state information - **Abstract:** We propose to use Tomlinson-Harashima Precoding (THP) for the reconfigurable intelligent surface (RIS)-aided multiple-input multiple-output (MIMO) broadcast channel where we assume a line of sight (LOS) connection between the base station (BS) and the RIS. In this scenario, nonlinear precoding, like THP or dirty paper coding (DPC), has certain advantages compared to linear precoding as it is more robust in case the BS-RIS channel is not orthogonal to the direct channel. Additionally, THP and DPC allow a simple phase shift optimization which is in strong contrast to linear precoding for which the solution is quite intricate. Besides being difficult to optimize, linear precoding has fundamental limitations when the phases are chosen randomly or based on statistical channel state information (CSI). These limitations do not hold for nonlinear precoding. Moreover, we show that the advantages of THP/DPC are especially pronounced for discrete phase shifts. ### HFE-RWKV: High-Frequency Enhanced RWKV Model for Efficient Left Ventricle Segmentation in Pediatric Echocardiograms ⚑ DE - **DOI:** [10.1109/icassp49660.2025.10888300](https://doi.org/10.1109/icassp49660.2025.10888300) - **Metadaten:** Erschienen: 2025-03-12 · S. 1-5 · OpenAlex seit 2025-03-13 - **DE-Institutionen:** Saarland University - **Autoren:** - Zi Ye — Institute of Intelligent Software,Guangzhou,China - Tianxiang Chen — University of Science and Technology of China (CN) - Ziyang Wang — University of Oxford (GB) - Hanwei Zhang — Saarland University (DE) - Lijun Zhang — Institute of Software (CN); University of Chinese Academy of Sciences (CN) - **Topics:** Medical Image Segmentation Techniques; Artificial Intelligence in Healthcare and Education; Advanced Neural Network Applications - **Keywords:** left ventricle segmentation; pediatric echocardiography; RWKV; high-frequency feature enhancement; computational efficiency - **Abstract:** Automated ventricular function analysis can improve healthcare in resource-scarce areas, but current segmentation methods struggle with accurately delineating the irregular shape of the left ventricle due to a lack of emphasis on exploring the high-frequency target boundary features, and computational inefficiency is another concern. To address the two challenges, we turn to a novel and efficient basic structure, RWKV, and propose High-Frequency Enhanced RWKV (HFE-RWKV) for accurate and efficient left ventricle segmentation. Specifically, we propose the HFE-RWKV block as the encoder’s core to augment the high-frequency component, which is also the boundary area of the left ventricles in pediatric echocardiograms. In this way, the target boundaries can be explored more adequately during feature extraction. We propose space-frequency consistency loss to refine the shape of predicted masks further. Specifically, our new loss function incorporates spatial and frequency domain loss components to jointly refine predicted mask shapes in cases where current spatial-domain segmentation losses cannot be optimized further. Experiments on two public datasets prove our HFE-RWKV’s superiority in accuracy and efficiency. Specifically, our HFE-RWKV outperforms U-Mamba [12] by 2% in Dice Similarity Coefficient (DSC) while using only 67% of the parameters and 26% of the computational complexity. The code is available at https://github.com/yezizi1022/HFE-RWKV. ### MutualForce: Mutual-Aware Enhancement for 4D Radar-LiDAR 3D Object Detection ⚑ DE - **DOI:** [10.1109/icassp49660.2025.10887748](https://doi.org/10.1109/icassp49660.2025.10887748) - **Metadaten:** Erschienen: 2025-03-12 · S. 1-5 · OpenAlex seit 2025-03-13 - **DE-Institutionen:** Technical University of Munich; Infineon Technologies (Germany) - **Autoren:** - Xiangyuan Peng — Technical University of Munich (DE) - Huawei Sun — Technical University of Munich (DE) - Kay Bierzynski — Infineon Technologies (Germany) (DE) - Anton Fischbacher — Technical University of Munich (DE) - Lorenzo Servadei — Technical University of Munich (DE) - Robert Wille — Technical University of Munich (DE) - **Topics:** Autonomous Vehicle Technology and Safety; Advanced Neural Network Applications; Radar Systems and Signal Processing - **Keywords:** 3D object detection; autonomous driving perception; weather robustness - **Abstract:** Radar and LiDAR have been widely used in autonomous driving as LiDAR provides rich structure information, and radar demonstrates high robustness under adverse weather. Recent studies highlight the effectiveness of fusing radar and LiDAR point clouds. However, challenges remain due to the modality misalignment and information loss during feature extractions. To address these issues, we propose a 4D radar-LiDAR framework to mutually enhance their representations. Initially, the indicative features from radar are utilized to guide both radar and LiDAR geometric feature learning. Subsequently, to mitigate their sparsity gap, the shape information from LiDAR is used to enrich radar BEV features. Extensive experiments on the View-of-Delft (VoD) dataset demonstrate our approach’s superiority over existing methods, achieving the highest mAP of 71.76% across the entire area and 86.36% within the driving corridor. Especially for cars, we improve the AP by 4.17% and 4.20% due to the strong indicative features and symmetric shapes. ### DoA-Aided MMSE Channel Estimation for Wireless Communication Systems ⚑ DE - **DOI:** [10.1109/icassp49660.2025.10890513](https://doi.org/10.1109/icassp49660.2025.10890513) - **Metadaten:** Erschienen: 2025-03-12 · S. 1-5 · OpenAlex seit 2025-03-13 - **DE-Institutionen:** Technical University of Munich - **Autoren:** - Franz Weißer — Technical University of Munich (DE) - Nurettin Turan — Technical University of Munich (DE) - Wolfgang Utschick — Technical University of Munich (DE) - **Topics:** Advanced Wireless Communication Techniques; Direction-of-Arrival Estimation Techniques; Millimeter-Wave Propagation and Modeling - **Keywords:** direction of arrival; MMSE channel estimation; side information; two-stage estimation; subspace decomposition; Gaussian mixture model; multipath propagation; computational complexity reduction - **Abstract:** This paper investigates using side information in minimum mean square error (MMSE) estimation. We propose a direction-of-arrival (DoA)-aided two-stage channel estimation technique that utilizes information about the dominant direction of the channel. To this end, the decomposition of the MMSE channel estimation into two orthogonal subspaces is formulated. After estimating the channel along the dominant direction, we utilize a Gaussian mixture model to estimate the conditionally Gaussian distributed random vector, which represents the multipath propagation. The proposed two-stage estimator allows pre-computing the respective estimation filters, tremendously reducing the computational complexity. Numerical simulations depict the superior performance of our proposed two-stage estimation approach compared to state-of-the-art methods. ### Factorized-VITS: Decoupling Prosody and Text in End-to-End Speech Synthesis without External or Secondary Aligner ⚑ DE - **DOI:** [10.1109/icassp49660.2025.10890003](https://doi.org/10.1109/icassp49660.2025.10890003) - **Metadaten:** Erschienen: 2025-03-12 · S. 1-5 · OpenAlex seit 2025-03-13 - **DE-Institutionen:** Karlsruhe Institute of Technology - **Autoren:** - Yining Liu — Karlsruhe Institute of Technology (DE) - Alexander H. Waibel — Karlsruhe Institute of Technology (DE) - **Topics:** Speech Recognition and Synthesis; Speech and Audio Processing; Voice and Speech Disorders - **Keywords:** text-to-speech; VITS; prosody modeling; in-context learning - **Abstract:** We propose Factorized-VITS, an advanced end-to-end text-to-speech model that incorporates explicit text-side prosody modeling control into VITS while achieving a clean factorization of the audio prior hidden space into text and prosody subspaces. Unlike previous works that rely on external or secondary aligners, Factorized-VITS is the first work attempting to do on-the-fly alignment in the factorized text subspace without introducing extra parameters, which not only simplifies the training procedure but also enables the use of a more complex prosody prior. Our experiments demonstrate the accuracy and effectiveness of this approximation strategy. Furthermore, we implement an in-context learning joint predictor for pitch, energy, and duration, which offers a flexible streaming deployment option. ### FlowMAC: Conditional Flow Matching for Audio Coding at Low Bit Rates ⚑ DE - **DOI:** [10.1109/icassp49660.2025.10888898](https://doi.org/10.1109/icassp49660.2025.10888898) - **Metadaten:** Erschienen: 2025-03-12 · S. 1-5 · OpenAlex seit 2025-03-13 - **DE-Institutionen:** Fraunhofer Institute for Integrated Circuits; International Audio Laboratories Erlangen - **Autoren:** - Nicola Pia — Fraunhofer Institute for Integrated Circuits (DE) - Martin Strauss — International Audio Laboratories Erlangen (DE) - Markus Multrus — Fraunhofer Institute for Integrated Circuits (DE) - Bernd Edler — International Audio Laboratories Erlangen (DE) - **Topics:** Speech and Audio Processing; Music and Audio Processing; Speech Recognition and Synthesis - **Keywords:** neural audio codecs; conditional flow matching; Mel spectrogram; continuous normalizing flows; ordinary differential equation solvers; CPU-only inference; perceptual audio quality - **Abstract:** This paper introduces FlowMAC, a novel neural audio codec for high-quality general audio compression at low bit rates based on conditional flow matching (CFM). FlowMAC jointly learns a mel spectrogram encoder, quantizer and decoder. At inference time the decoder integrates a continuous normalizing flow via an ODE solver to generate a high-quality mel spectrogram. This is the first time that a CFM-based approach is applied to general audio coding, enabling a scalable, simple and memory efficient training. Our subjective evaluations show that FlowMAC at 3 kbps achieves similar quality as state-of-the-art GAN-based and DDPM-based neural audio codecs at double the bit rate. Moreover, FlowMAC offers a tunable inference pipeline, which permits to trade off complexity and quality. This enables real-time coding on CPU, while maintaining high perceptual quality. ### MPAM-3DGS: Multi-Parametric Adversarial Manipulation for 3D Gaussian Splatting ⚑ DE - **DOI:** [10.1109/icassp49660.2025.10889677](https://doi.org/10.1109/icassp49660.2025.10889677) - **Metadaten:** Erschienen: 2025-03-12 · S. 1-5 · OpenAlex seit 2025-03-13 - **DE-Institutionen:** Saarland University - **Autoren:** - Wenxiang Jiang — Ocean University of China (CN) - Hanwei Zhang — Saarland University (DE) - Weigang Wang — Ocean University of China (CN) - Zhongwen Guo — Ocean University of China (CN) - Tianao Zhang — Ocean University of China (CN) - Hao Wang — Xidian University (CN) - **Topics:** Adversarial Robustness in Machine Learning; 3D Shape Modeling and Analysis; Security and Verification in Computing - **Keywords:** 3D Gaussian Splatting; adversarial robustness; adversarial attacks; parameter perturbation; 3D object detection; 3D classification; downstream tasks - **Abstract:** 3D Gaussian Splatting (3DGS) is gaining popularity in fields such as robotics, autonomous driving, and virtual reality, due to its effectiveness and efficiency. Given that some tasks involve high risks, it is crucial to investigate the adversarial robustness of 3DGS and its downstream tasks—a topic that remains largely unexplored. In this study, we introduce a framework, Multi-Parametric Adversarial Manipulation for 3D Gaussian Splatting (MPAM-3DGS), that allows to attack 3DGS and its downstream tasks, such as object detection and classification, by perturbing a specified subset of parameters. Leveraging this framework, we examine the adversarial sensitivity of each 3DGS parameter and propose two strategies to attack multiple parameters based on our observations. To our knowledge, this is the first study to explore the adversarial robustness of 3DGS. Our experimental results demonstrate the effectiveness of our attacks on downstream tasks and the invisibility of perturbations in 3DGS. The code can be found at https://github.com/jiang-wenxiang/MPAM-3DGS. ### LIMMITS’25: Multilingual Streaming TTS With Neural Codecs for Indian Languages ⚑ DE - **DOI:** [10.1109/icassp49660.2025.10890595](https://doi.org/10.1109/icassp49660.2025.10890595) - **Metadaten:** Erschienen: 2025-03-12 · S. 1-2 · OpenAlex seit 2025-03-13 - **DE-Institutionen:** Deutsche Gesellschaft für Internationale Zusammenarbeit - **Autoren:** - Philipp Olbrich — Deutsche Gesellschaft für Internationale Zusammenarbeit (PH); Deutsche Gesellschaft für Internationale Zusammenarbeit (DE) - Hema A. Murthy — Indian Institute of Technology Madras (IN) - Pranaw Kumar — CDAC,Mumbai,India - Shinji Watanabe — Carnegie Mellon University (US) - Sheng Zhao — Microsoft (United States) (US) - Mark Hasegawa‐Johnson — University of Illinois Urbana-Champaign (US) - **Topics:** Speech Recognition and Synthesis; Speech and Audio Processing; Voice and Speech Disorders - **Keywords:** neural codecs; multilingual speech synthesis; Indian languages; Gujarati; Bhojpuri; Kannada language; Indian English; speech naturalness - **Abstract:** This work provides a summary of the Multilingual streaming TTS with neural codecs for Indian languages challenge (LIMMITS’25), organized as part of the ICASSP 2025 signal processing grand challenge. Towards this, 278 hours of TTS data in 4 Indian languages - Gujarati, Indian English, Bhojpuri, and Kannada got released. The challenge focuses on advancing research in neural codec-based and streaming TTS systems. The top teams in the challenge attained high subjective scores on naturalness and similarity, thus contributing to the progress in text-to-speech generation systems. ### On the Relation Between Speech Quality and Quantized Latent Representations of Neural Codecs ⚑ DE - **DOI:** [10.1109/icassp49660.2025.10890357](https://doi.org/10.1109/icassp49660.2025.10890357) - **Metadaten:** Erschienen: 2025-03-12 · S. 1-5 · OpenAlex seit 2025-03-12 - **DE-Institutionen:** Fraunhofer Institute for Integrated Circuits - **Autoren:** - Mhd Modar Halimeh — Fraunhofer Institute for Integrated Circuits (DE) - Matteo Torcoli — Fraunhofer Institute for Integrated Circuits (DE) - Philipp Grundhuber — Fraunhofer Institute for Integrated Circuits (DE) - Emanuël A. P. Habets — Fraunhofer Institute for Integrated Circuits (DE) - **Topics:** Speech and Audio Processing; Image and Video Quality Assessment; Speech Recognition and Synthesis - **Keywords:** neural audio codecs; speech quality assessment; latent representations; quantization error; subjective speech quality; low bit rate audio coding; speech quality prediction - **Abstract:** Neural audio signal codecs have attracted significant attention in recent years. In essence, the impressive low bitrate achieved by such encoders is enabled by learning an abstract representation that captures the properties of encoded signals, e.g., speech. In this work, we investigate the relation between the latent representation of the input signal learned by a neural codec and the quality of speech signals. To do so, we introduce Latent-representation-to-Quantization error Ratio (LQR) measures, which quantify the distance from the idealized neural codec’s speech signal model for a given speech signal. We compare the proposed metrics to intrusive measures as well as data-driven supervised methods using two subjective speech quality datasets. This analysis shows that the proposed LQR correlates strongly (up to 0.9 Pearson’s correlation) with the subjective quality of speech. Despite being a non-intrusive metric, this yields a competitive performance with, or even better than, other pre-trained and intrusive measures. These results show that LQR is a promising basis for more sophisticated speech quality measures. ### Cross-Channel Unlabeled Sensing over a Union of Signal Subspaces ⚑ DE - **DOI:** [10.1109/icassp49660.2025.10888212](https://doi.org/10.1109/icassp49660.2025.10888212) - **Metadaten:** Erschienen: 2025-03-12 · S. 1-5 · OpenAlex seit 2025-03-13 - **DE-Institutionen:** Technische Universität Darmstadt - **Autoren:** - Taulant Koka — Technische Universität Darmstadt (DE) - Manolis C. Tsakiris — Chinese Academy of Sciences (CN); Academy of Mathematics and Systems Science (CN) - Benjamín Béjar — Paul Scherrer Institute (CH); Swiss Data Science Center (CH) - Michael Muma — Technische Universität Darmstadt (DE) - **Topics:** Sparse and Compressive Sensing Techniques; Mathematical Analysis and Transform Methods; Distributed Sensor Networks and Detection Algorithms - **Keywords:** union of subspaces; compressed sensing; whole-brain calcium imaging; multi-target tracking - **Abstract:** Cross-channel unlabeled sensing addresses the problem of recovering a multi-channel signal from measurements that were shuffled across channels. This work expands the cross-channel unlabeled sensing framework to signals that lie in a union of subspaces. The extension allows for handling more complex signal structures and broadens the framework to tasks like compressed sensing. These mismatches between samples and channels often arise in applications such as whole-brain calcium imaging of freely moving organisms or multi-target tracking. We improve over previous models by deriving tighter bounds on the required number of samples for unique reconstruction, while supporting more general signal types. The approach is validated through an application in whole-brain calcium imaging, where organism movements disrupt sample-to-neuron mappings. This demonstrates the utility of our framework in real-world settings with imprecise sample-channel associations, achieving accurate signal reconstruction. ## Advanced Electronic Materials (AEM) — 1 neu · _materials_ ### Two‐Terminal Synaptic Devices Based on a Single‐Component Conjugated Polyelectrolyte for Noise‐Tolerant Pattern Encoding ⚑ DE - **DOI:** [10.1002/aelm.70640](https://doi.org/10.1002/aelm.70640) - **Metadaten:** Erschienen: 2026-11-15 · OpenAlex seit 2026-11-16 - **DE-Institutionen:** Humboldt-Universität zu Berlin; University of Wuppertal; Helmholtz-Zentrum Berlin für Materialien und Energie - **Autoren:** - Fabian Gärisch — Humboldt-Universität zu Berlin (DE) - Ullrich Scherf — University of Wuppertal (DE) - Hauke Jan Jötten — University of Wuppertal (DE) - Emil J. W. List — Helmholtz-Zentrum Berlin für Materialien und Energie (DE); Humboldt-Universität zu Berlin (DE) - Giovanni Ligorio — Humboldt-Universität zu Berlin (DE) - **Topics:** Advanced Memory and Neural Computing; Conducting polymers and applications; Organic Electronics and Photovoltaics - **Abstract:** ABSTRACT Neuromorphic devices, inspired by the efficiency of biological neural networks, hold promise for local, low‐power processing in artificial intelligence and edge‐computing applications. Organic materials are particularly attractive for such devices due to their chemical tunability, enabling precise control of ion transport dynamics critical for synaptic emulation. Here we present an artificial two‐terminal synaptic device based on a single‐component conjugated polythiophene‐based polyelectrolyte. The single‐component design avoids inter‐component phase separation; the devices additionally exhibit stable and reproducible operation. Electrical characterization, including impedance spectroscopy, reveals ion‐mediated transport and electrochemical doping processes governing device operation. The devices exhibit key neuromorphic properties such as paired‐pulse facilitation with fast response times, pronounced signal amplification, and stable conductance modulation under pulsed operation. Pattern‐encoding experiments with a 16‐device array show that four patterns remain correctly classified up to 33.3% input noise. These results highlight the potential of single‐component organic materials for scalable, high‐performance, and noise‐resilient neuromorphic computing platforms.