# Neue Publikationen in Ziel-Venues – 2026-10-12 Neue Werke: 28 · davon mit DE-Affiliation: 28 Venues: 60 · abgefragt ab publication_date 2026-09-12 · letzter Lauf: 2026-10-12T18:10:38 ## IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) — 27 neu · _conference_ ### Acoustic Teleportation Via Disentangled Neural Audio Codec Representations ⚑ DE - **DOI:** [10.1109/icassp55912.2026.11463294](https://doi.org/10.1109/icassp55912.2026.11463294) - **Metadaten:** Erschienen: 2026-04-21 · S. 14442-14446 · OpenAlex seit 2025-10-17 - **DE-Institutionen:** Fraunhofer Institute for Integrated Circuits; International Audio Laboratories Erlangen - **Autoren:** - Philipp Grundhuber — Fraunhofer Institute for Integrated Circuits (DE) - Mhd Modar Halimeh — Fraunhofer Institute for Integrated Circuits (DE) - Emanuël A. P. Habets — International Audio Laboratories Erlangen (DE) - **Topics:** Speech and Audio Processing; Speech Recognition and Synthesis; Music and Audio Processing - **Keywords:** disentangled representations; speech dereverberation; acoustic embeddings; speaker identity preservation; virtual acoustic environments - **Abstract:** This paper presents an approach for acoustic teleportation by disentangling speech content from acoustic environment characteristics in neural audio codec representations. Acoustic teleportation transfers room characteristics between speech recordings while preserving content and speaker identity, enabling applications in telecommunications and virtual acoustic environments. Extending prior work by adopting the EnCodec architecture, we achieve substantial objective quality improvements with non-intrusive ScoreQ scores of 3.03, compared to 2.44 for previous methods. Our training strategy incorporates five tasks: clean reconstruction, reverberated reconstruction, dereverberation, and two variants of acoustic teleportation. We analyze the trade-off between temporal downsampling of acoustic embeddings and reconstruction quality, demonstrating that even a downsampling factor of two yields a statistically significant degradation in reconstruction quality. The learned acoustic embeddings exhibit a strong correlation with reverberation time. t-SNE analysis reveals that acoustic embeddings cluster predominantly by room while speech embeddings cluster by speaker, confirming substantial disentanglement sufficient for effective acoustic teleportation. ### Fast-Slow LoRa: Achieving Efficient Continual Learning via Fast-Slow Learning and Redundancy Pruning ⚑ DE - **DOI:** [10.1109/icassp55912.2026.11463985](https://doi.org/10.1109/icassp55912.2026.11463985) - **Metadaten:** Erschienen: 2026-04-21 · S. 4576-4580 · OpenAlex seit 2026-04-22 - **DE-Institutionen:** University of Stuttgart - **Autoren:** - Chenrui Xiong — University of Electronic Science and Technology of China (CN) - Lin Zhang — University of Stuttgart (DE) - Zhao Kang — University of Electronic Science and Technology of China (CN) - **Topics:** Domain Adaptation and Few-Shot Learning; Advanced Neural Network Applications; Topic Modeling - **Keywords:** continual learning; catastrophic forgetting; LoRa; parameter-efficient fine-tuning - **Abstract:** The core challenge of continual learning (CL) is catastrophic forgetting—models lose prior knowledge when learning new tasks. While PEFT methods like LoRA reduce adaptation costs, they suffer from parameter interference due to poor task isolation. We propose Fast-Slow LoRA (FSLoRA), a novel CL framework that decomposes LoRA into two task-agnostic expert pools: a Slow Expert Pool (SEP) for stable, cross-task representations, and a dynamic Fast Expert Pool (FEP) for task-specific features, jointly routed by a unified gate. To further curb redundancy and negative transfer, we introduce an Expert Pruning Mechanism (EPM) that removes over-lapping experts via semantic similarity and contribution scores. FSLoRA achieves state-of-the-art performance on no-replay CIL benchmarks, delivering superior accuracy with minimal parameter overhead. ### Differentiable Grouped Feedback Delay Networks for Learning Direction and Position-Dependent Late Reverberation ⚑ DE - **DOI:** [10.1109/icassp55912.2026.11461624](https://doi.org/10.1109/icassp55912.2026.11461624) - **Metadaten:** Erschienen: 2026-04-21 · S. 14567-14571 · OpenAlex seit 2026-04-22 - **DE-Institutionen:** Friedrich-Alexander-Universität Erlangen-Nürnberg - **Autoren:** - Orchisama Das — King's College London (GB) - Sebastian J. Schlecht — Friedrich-Alexander-Universität Erlangen-Nürnberg (DE) - Gloria Dal Santo — Aalto University (FI) - Zoran Cvetković — King's College London (GB) - **Topics:** Speech and Audio Processing; Music Technology and Sound Studies; Music and Audio Processing - **Keywords:** late reverberation; binaural rendering - **Abstract:** Late reverberation in coupled spaces shows direction and position-dependent decay. Grouped Feedback Delay Networks (GFDNs) are artificial reverberators capable of efficient rendering of coupled room acoustics by modelling multi-slope decays. We extend the concept of Differentiable GFDN (DiffGFDN) to synthesise position- and direction-dependent late reverberation. The proposed DiffGFDNs operate in octave bands, with spherical harmonic receiver gains predicted from listener positions via a Multi-Layer Perceptron. Training minimises spectral deviation, promotes low sparsity, and matches Directional Energy Decay Curves (DEDCs). At inference, signals are processed in frequency subbands, beamformed, and rendered binaurally or over loudspeakers. The proposed method shows slightly higher DEDC errors (+1.5 dB at 0.6 m, +0.9 dB at 0.9 m) than the best baseline, but is the most computationally efficient, enabling faster 6-DoF rendering than time-space-varying convolution. ### Sparsity-based learning enables absorption contrast of mixed-contrast objects in tomographic X-ray near-field holography ⚑ DE - **DOI:** [10.1109/icassp55912.2026.11461296](https://doi.org/10.1109/icassp55912.2026.11461296) - **Metadaten:** Erschienen: 2026-04-21 · S. 21937-21941 · OpenAlex seit 2026-04-22 - **DE-Institutionen:** Deutsches Elektronen-Synchrotron DESY - **Autoren:** - Johannes Dora — Deutsches Elektronen-Synchrotron DESY (DE) - Christian G. Schroer — Deutsches Elektronen-Synchrotron DESY (DE) - Johannes Hagemann — Deutsches Elektronen-Synchrotron DESY (DE) - **Topics:** Advanced X-ray Imaging Techniques; Digital Holography and Microscopy; X-ray Diffraction in Crystallography - **Keywords:** tomographic reconstruction; complex refractive index; imaginary part of refractive index; wavelet coefficients; absorption contrast; hard X-rays - **Abstract:** In the hard X-ray regime, the imaginary part β of the complex refractive index of matter is usually hundred times smaller than the real part δ and has a low signal-to-noise ratio in near-field holograms. Commonly used algorithms that implicitly or directly reconstruct the complex refractive indices of mixed-contrast objects typically fail to recover the imaginary part, without imposing strong priors on β. In this article, we present a learning-based approach to extract wavelet co-efficients of absorption images from single holograms. We demonstrate the robustness of the approach by reconstructing an absorption tomogram from experimental data. ### Bone-Conduction Guided Multimodal Speech Enhancement with Conditional Diffusion Models ⚑ DE - **DOI:** [10.1109/icassp55912.2026.11463781](https://doi.org/10.1109/icassp55912.2026.11463781) - **Metadaten:** Erschienen: 2026-04-21 · S. 14387-14391 · OpenAlex seit 2026-04-22 - **DE-Institutionen:** Universität Hamburg - **Autoren:** - Sina Khanagha — Universität Hamburg (DE) - Bunlong Lay — Universität Hamburg (DE) - Timo Gerkmann — Universität Hamburg (DE) - **Topics:** Speech and Audio Processing; Speech Recognition and Synthesis; Image Enhancement Techniques - **Keywords:** single-channel speech enhancement; conditional diffusion models; speech enhancement - **Abstract:** Single-channel speech enhancement models face significant performance degradation in extremely noisy environments. While prior work has shown that complementary bone-conducted speech can guide enhancement, effective integration of this noise-immune modality remains a challenge. This paper introduces a novel multimodal speech enhancement framework that integrates bone-conduction sensors with air-conducted microphones using a conditional diffusion model. Our proposed model significantly outperforms previously established multimodal techniques and a powerful diffusion-based single-modal baseline across a wide range of acoustic conditions1. ### A Comparative Study on how Data Normalization Affects Zero-Shot Generalization in Time Series Foundation Models ⚑ DE - **DOI:** [10.1109/icassp55912.2026.11464422](https://doi.org/10.1109/icassp55912.2026.11464422) - **Metadaten:** Erschienen: 2026-04-21 · S. 3436-3440 · OpenAlex seit 2025-12-04 - **DE-Institutionen:** Siemens (Germany); Ludwig-Maximilians-Universität München - **Autoren:** - Ihab Ahmed — Siemens (Germany) (DE) - Denis Krompaß — Siemens (Germany) (DE) - Cheng Feng — Siemens (Germany) (DE) - Volker Tresp — Ludwig-Maximilians-Universität München (DE) - **Topics:** Time Series Analysis and Forecasting; Forecasting Techniques and Applications; Stock Market Forecasting Methods - **Keywords:** time series foundation models; input normalization; zero-shot generalization; scale variation; point forecasting; accuracy-efficiency trade-off - **Abstract:** We investigate input normalization methods for Time Series Foundation Models (TSFMs). While normalization is well-studied in dataset-specific time series models, it remains overlooked in TSFMs where generalization is critical. Time series data, unlike text or images, exhibits significant scale variation across domains and channels, coupled with non-stationarity, can undermine TSFM performance regardless of architectural complexity. Through systematic evaluation using four architecturally diverse TSFMs, we empirically establish RevIN as the most efficient approach, reducing zero-shot MASE by 89% relative to raw baseline and by 44% in comparison to other normalization methods, while matching the best in-domain accuracy without any dataset-level preprocessing—yielding the highest accuracy-efficiency trade-off. Yet its effective utilization depends on architectural design choices and optimization objective, particularly with respect to training loss scale sensitivity and model type (point-forecast-, distribution- or LLM-based models). ### 3D-DOPAMIN: Diffusion-Optimized Projection Algorithm for 3D Multi-Coil MRI Reconstruction ⚑ DE - **DOI:** [10.1109/icassp55912.2026.11462255](https://doi.org/10.1109/icassp55912.2026.11462255) - **Metadaten:** Erschienen: 2026-04-21 · S. 7056-7060 · OpenAlex seit 2026-04-22 - **DE-Institutionen:** Forschungszentrum Jülich - **Autoren:** - Arya Bangun — Forschungszentrum Jülich (DE) - Alessio Quercia — Forschungszentrum Jülich (DE) - Lena Krieger — Forschungszentrum Jülich (DE) - Zhuo Cao — Forschungszentrum Jülich (DE) - Xuan Zhao — Forschungszentrum Jülich (DE) - Hanno Scharr — Forschungszentrum Jülich (DE) - **Topics:** Advanced MRI Techniques and Applications; Image and Signal Denoising Methods; Medical Imaging Techniques and Applications - **Keywords:** 3D MRI reconstruction; multi-coil MRI; parallel imaging; diffusion-weighted imaging ### Single-DMRS Based CFO Estimation for Low Latency Cellular Communications ⚑ DE - **DOI:** [10.1109/icassp55912.2026.11464821](https://doi.org/10.1109/icassp55912.2026.11464821) - **Metadaten:** Erschienen: 2026-04-21 · S. 21486-21490 · OpenAlex seit 2026-04-22 - **DE-Institutionen:** Huawei German Research Center - **Autoren:** - Ahmed Abdelkader — Huawei German Research Center (DE) - Waqar Anwar — Huawei German Research Center (DE) - Zhibin Yu — Huawei German Research Center (DE) - Xiaofeng Wu — Huawei German Research Center (DE) - **Topics:** Advanced Wireless Communication Techniques; PAPR reduction in OFDM; Advanced MIMO Systems Optimization - **Keywords:** carrier frequency offset estimation; ultra-reliable low-latency communication; 5G NR; pilot overhead reduction; Cramer-Rao lower bound; error vector magnitude; high-order modulation - **Abstract:** The emerging cellular communication systems require both high throughput and low latency, which can be realized by high-order modulation schemes and reliable carrier-frequency offset (CFO) estimation with a low pilot overhead. In the fifth generation (5G) new radio (NR) system, single front-loaded demodulation reference symbols (DMRS) have been introduced for low-latency channel estimation in ultra reliable low latency communications (URLLC) scenarios. However, due to the multi-access interference (MAI) within a single DMRS, the CFO estimation still has to rely on other paired reference symbols, such as the tracking reference symbols (TRS). This paper proposes a novel method which enables CFO estimation by using a single DMRS. The method leverages the structure of the block interleaved frequency division multiple access (B-IFDMA) to perform joint MAI separation and CFO estimation. Simulations show that the proposed method can approach the Cramer-Rao lower bound (CRLB) and can sufficiently avoid (DL) error vector magnitude (EVM) degradation for high-order modulation based low-latency communications. ### Cooperative Multi-Modal Navigation for Heterogeneous Marine Robots ⚑ DE - **DOI:** [10.1109/icassp55912.2026.11461988](https://doi.org/10.1109/icassp55912.2026.11461988) - **Metadaten:** Erschienen: 2026-04-21 · S. 22152-22156 · OpenAlex seit 2026-04-22 - **DE-Institutionen:** Deutsches Zentrum für Luft- und Raumfahrt e. V. (DLR) - **Autoren:** - Maureen Petersilka — Deutsches Zentrum für Luft- und Raumfahrt e. V. (DLR) (DE) - Armin Dammann — Deutsches Zentrum für Luft- und Raumfahrt e. V. (DLR) (DE) - Siwei Zhang — Deutsches Zentrum für Luft- und Raumfahrt e. V. (DLR) (DE) - Fabio Broghammer — Deutsches Zentrum für Luft- und Raumfahrt e. V. (DLR) (DE) - Thomas Strang — Deutsches Zentrum für Luft- und Raumfahrt e. V. (DLR) (DE) - Michael Walter — Deutsches Zentrum für Luft- und Raumfahrt e. V. (DLR) (DE) - **Topics:** Maritime Navigation and Safety; Underwater Vehicles and Communication Systems; Distributed Control Multi-Agent Systems - **Keywords:** cooperative navigation; underwater robots; multi-robot coordination; sensor fusion ### DisContSE: Single-Step Diffusion Speech Enhancement based on Joint Discrete and Continuous Embeddings ⚑ DE - **DOI:** [10.1109/icassp55912.2026.11460808](https://doi.org/10.1109/icassp55912.2026.11460808) - **Metadaten:** Erschienen: 2026-04-21 · S. 17392-17396 · OpenAlex seit 2026-04-22 - **DE-Institutionen:** Technische Universität Braunschweig - **Autoren:** - Yihui Fu — Technische Universität Braunschweig (DE) - Tim Fingscheidt — Technische Universität Braunschweig (DE) - **Topics:** Speech and Audio Processing; Speech Recognition and Synthesis; Image and Signal Denoising Methods - **Keywords:** speech enhancement; diffusion models; continuous embeddings ### 2025 Urgent Speech Enhancement Challenge Multilingual P.808 Listening Tests: Approach and Results ⚑ DE - **DOI:** [10.1109/icassp55912.2026.11465025](https://doi.org/10.1109/icassp55912.2026.11465025) - **Metadaten:** Erschienen: 2026-04-21 · S. 22477-22481 · OpenAlex seit 2026-04-22 - **DE-Institutionen:** Technische Universität Braunschweig - **Autoren:** - Marvin Sach — Technische Universität Braunschweig (DE) - Yihui Fu — Technische Universität Braunschweig (DE) - Kohei Saijo — Waseda University (JP) - Wangyou Zhang — Shanghai Jiao Tong University (CN) - Samuele Cornell — Carnegie Mellon University (US) - Robin Scheibler — Google DeepMind,Japan - C. Li — Shanghai Jiao Tong University (CN) - Zhaoheng Ni — Meta (United States) (US) - Anurag Kumar — Meta (United States) (US) - Wei Wang — Shanghai Jiao Tong University (CN) - Yanmin Qian — Shanghai Jiao Tong University (CN) - Shinji Watanabe — Carnegie Mellon University (US) - Tim Fingscheidt — Technische Universität Braunschweig (DE) - **Topics:** Speech and Audio Processing; Speech Recognition and Synthesis; Image and Video Quality Assessment - **Keywords:** speech enhancement; subjective listening tests; absolute category rating; hallucination detection; reproducible evaluation - **Abstract:** In speech quality estimation for speech enhancement (SE) systems, subjective listening tests so far are considered as the gold standard. This should be even more true considering the large influx of new generative or hybrid methods into the field, revealing issues of some objective metrics. Efforts such as the Interspeech 2025 URGENT Speech Enhancement Challenge also involving non-English datasets add the aspect of multilinguality to the testing procedure. In this paper, we provide updated challenge results on the full multilingual test set and surprising insights into URGENT Challenge results, questioning the reliability of (P.808) absolute category rating (ACR) subjective testing as gold standard in the age of generative AI. Particularly, it seems that for generative SE methods, subjective (ACR MOS) and objective (DNSMOS, NISQA) reference-free metrics should be accompanied by objective phone fidelity metrics to reliably detect hallucinations. We briefly recap the ITU-T P.808 crowdsourced listening test method and detail our proposed and employed localization protocol for text and audio in subjective ACR tests, that we open-source, enabling reproducible evaluation of multilingual speech processing tasks. ### 3GPP IVAS Codec – Perspectives on Development, Testing and Standardization ⚑ DE - **DOI:** [10.1109/icassp49660.2025.10889939](https://doi.org/10.1109/icassp49660.2025.10889939) - **Metadaten:** Erschienen: 2025-03-12 · S. 1-5 · OpenAlex seit 2025-10-10 - **DE-Institutionen:** Fraunhofer Institute for Integrated Circuits - **Autoren:** - Stefan Bruhn — Dolby - Tomas Toftgard — Ericsson (Sweden) (SE) - Stefan Döhla — Fraunhofer Institute for Integrated Circuits (DE) - H-Y. Su — Huawei Technologies (China) (CN) - Laura Laaksonen — Nokia - T. Moriya — NTT (Japan) (JP) - Stéphane Ragot — Orange (France) (FR) - H. Ehara — Panasonic (Japan) (JP) - Marek Szczerba — Philips (Netherlands) (NL) - Имре Вaргa — keine Affiliation - A. Schevciw — Qualcomm (United States) (US) - Milan Jelinek — VoiceAge (Canada) (CA) - **Topics:** Speech and Audio Processing; Telecommunications and Broadcasting Technologies; Speech Recognition and Synthesis - **Keywords:** 3GPP; spatial audio coding; spatial audio; mobile communication - **Abstract:** The standardization of the codec for Immersive Voice and Audio Services (IVAS) was completed by the 3rd Generation Partnership Project in June 2024. The IVAS codec goes beyond traditional mono voice coding by representing and reproducing the spatial characteristics of sound, creating an immersive auditory experience. It opens the space for new applications in the realm of mobile communications and user-generated live content streaming such as immersive telephony and extended reality (XR) teleconferencing. The present paper provides a brief overview of the IVAS standard framework covering key features and properties and unique perspectives that are essential for understanding the underlying development and standardization processes that have led to this new 3GPP standard. ### Learning from Ambiguous Data with Hard Labels ⚑ DE - **DOI:** [10.1109/icassp49660.2025.10889047](https://doi.org/10.1109/icassp49660.2025.10889047) - **Metadaten:** Erschienen: 2025-03-12 · S. 1-5 · OpenAlex seit 2025-10-10 - **DE-Institutionen:** University of Tübingen - **Autoren:** - Zeke Xie — The Hong Kong University of Science and Technology (Guangzhou) (CN) - Zheng He — University of British Columbia (CA) - Nan Lu — University of Tübingen (DE) - Lichen Bai — The Hong Kong University of Science and Technology (Guangzhou) (CN) - Bao Li — The Hong Kong University of Science and Technology (Guangzhou) (CN) - Shuo Yang — Harbin Institute of Technology (CN) - Mingming Sun — Beijing Institute of Mathematical Sciences and Applications - Ping Li — VecML Inc.,Seattle,Washington,U.S.A - **Topics:** Machine Learning and Data Classification; Domain Adaptation and Few-Shot Learning; Adversarial Robustness in Machine Learning - **Keywords:** ambiguous data; soft labels; overconfidence; model generalization; label noise - **Abstract:** Real-world data often contains intrinsic ambiguity that the common single-hard-label annotation paradigm ignores. Standard training using ambiguous data with these hard labels may produce overly confident models and thus leading to poor generalization. In this paper, we propose a novel framework called Quantized Label Learning (QLL) to alleviate this issue. First, we formulate QLL as learning from (very) ambiguous data with hard labels: ideally, each ambiguous instance should be associated with a ground-truth soft-label distribution describing its corresponding probabilistic weight in each class, however, this is usually not accessible; in practice, we can only observe a quantized label, i.e., a hard label sampled (quantized) from the corresponding ground-truth soft-label distribution, of each instance, which can be seen as a biased approximation of the ground-truth soft-label. Second, we propose a Class-wise Positive-Unlabeled (CPU) risk estimator that allows us to train accurate classifiers from only ambiguous data with quantized labels. Third, to simulate ambiguous datasets with quantized labels in the real world, we design a mixing-based ambiguous data generation procedure for empirical evaluation. Experiments demonstrate that our CPU method can significantly improve model generalization performance and outperform the baselines. ### Explaining Representations in Correlation-based Deep Multiview Representation Learning ⚑ DE - **DOI:** [10.1109/icassp49660.2025.10888309](https://doi.org/10.1109/icassp49660.2025.10888309) - **Metadaten:** Erschienen: 2025-03-12 · S. 1-5 · OpenAlex seit 2025-10-10 - **DE-Institutionen:** Paderborn University - **Autoren:** - Maurice Kuschel — Paderborn University (DE) - Amr Alkhatib — KTH Royal Institute of Technology (SE) - Tanuj Hasija — Paderborn University (DE) - Henrik Boström — KTH Royal Institute of Technology (SE) - **Topics:** Explainable Artificial Intelligence (XAI); Topological and Geometric Data Analysis; Time Series Analysis and Forecasting - **Keywords:** deep canonical correlation analysis; explainable AI; SHAP; latent space interpretability; correlation maximization; medical time series; synthetic images - **Abstract:** Multiview representation learning techniques based on deep correlation maximization have become increasingly popular for learning meaningful and compact representations from multiview data. Even though their performance is state-of-the-art in many interpretability-critical fields, their black-box behavior poses a problem and restricts their usability. To overcome this restriction, we propose XDCCA (eXplanations for Deep Canonical Correlation Analysis), an explanation strategy using characteristic rules in combination with SHAP that exploits the inherent structure of latent spaces created by correlation maximization techniques. We demonstrate how XDCCA allows for interpreting learned representations and their correlation using real medical time series and synthetic image data. ### Dual-Function Waveform Design in Wireless Sensor Networks via SoS Optimization ⚑ DE - **DOI:** [10.1109/icassp49660.2025.10888149](https://doi.org/10.1109/icassp49660.2025.10888149) - **Metadaten:** Erschienen: 2025-03-12 · S. 1-5 · OpenAlex seit 2025-10-10 - **DE-Institutionen:** Valeo (Germany) - **Autoren:** - Robin Amar — University of Luxembourg (LU); Interdisciplinary Centre for Security, Reliability and Trust (LU) - Linlong Wu — University of Luxembourg (LU); Interdisciplinary Centre for Security, Reliability and Trust (LU) - Saeid Sedighi — Valeo (Germany) (DE) - Mohammad Alaee-Keerahroodi — University of Luxembourg (LU); Interdisciplinary Centre for Security, Reliability and Trust (LU) - Bhavani Shankar M. R. — University of Luxembourg (LU); Interdisciplinary Centre for Security, Reliability and Trust (LU) - **Topics:** Radar Systems and Signal Processing; Distributed Sensor Networks and Detection Algorithms; Ultra-Wideband Communications Technology - **Keywords:** integrated sensing and communication; waveform design; wireless sensor networks; sum-of-squares programming; coherent processing interval; quadratic phase; dynamic target detection; interference minimization - **Abstract:** Environmental sensing can be achieved using Wireless Sensor Networks (WSNs) with low-complexity sensors. Integrated sensing and communication (ISAC) allows simultaneous sensing and data transmission through optimized waveform, architecture, and protocol design. Given the multiple sensors in WSNs, minimizing interference and ensuring Doppler-tolerant pulses for detecting dynamic targets is crucial. This paper proposes a Doppler-tolerant waveform design that embeds multiple communication symbols within each pulse of a coherent processing interval (CPI). We formulate a Sum-of-Squares (SoS) optimization problem to minimize the cross-correlation integrated sidelobe level (CISL) for a MIMO sensor arrangement, constrained by unwrapped quadratic phase behavior. Simulation results confirm the effectiveness of the proposed approach. ### HypCAD: Geometry-Enhanced Hyperbolic Contrastive Learning for CAD Model Retrieval ⚑ DE - **DOI:** [10.1109/icassp49660.2025.10887722](https://doi.org/10.1109/icassp49660.2025.10887722) - **Metadaten:** Erschienen: 2025-03-12 · S. 1-5 · OpenAlex seit 2025-10-10 - **DE-Institutionen:** Siemens (Germany); Technical University of Munich - **Autoren:** - Adam Misik — Siemens (Germany) (DE) - Driton Salihu — Technical University of Munich (DE) - Xin Su — Technical University of Munich (DE) - Heike Brock — Siemens (Germany) (DE) - Eckehard G. Steinbach — Technical University of Munich (DE) - **Topics:** 3D Shape Modeling and Analysis; Advanced Image and Video Retrieval Techniques; Image Processing and 3D Reconstruction - **Keywords:** CAD model retrieval; hyperbolic space; contrastive learning; ShapeNet; mixed reality - **Abstract:** Retrieving CAD models for real-world object scans enhances object-level mapping, providing a nuanced spatial understanding crucial for precise interactions in robotics or mixed reality. Commonly, CAD model retrieval is performed by matching features learned in Euclidean space. However, learning discriminative features in Euclidean space faces significant challenges, primarily due to its flat nature and the wide variety of CAD models with different levels of detail. To address the limitations of Euclidean space and improve CAD model retrieval, this paper introduces HypCAD, a contrastive learning framework in hyperbolic space. We present a novel geometry-enhanced hyperbolic distance and utilize a three-component contrastive learning loss to learn hyperbolic feature representations for the CAD model retrieval task. We demonstrate HypCAD’s superior retrieval accuracy through comparisons with baseline contrastive learning methods on both the synthetic ShapeNet dataset and the real-world Scan2CAD dataset. ### A spectrum-enhanced attention model for semantic segmentation of remote sensing images ⚑ DE - **DOI:** [10.1109/icassp49660.2025.10887646](https://doi.org/10.1109/icassp49660.2025.10887646) - **Metadaten:** Erschienen: 2025-03-12 · S. 1-5 · OpenAlex seit 2025-10-10 - **DE-Institutionen:** Friedrich-Alexander-Universität Erlangen-Nürnberg - **Autoren:** - Xin Li — Hohai University (CN) - Feng Xu — Hohai University (CN) - Feifei Tao — Hohai University (CN) - Yao Tong — Nanjing University of Chinese Medicine (CN) - Xin Lyu — Hohai University (CN) - Jianyi Zhong — Hohai University (CN) - André Kaup — Friedrich-Alexander-Universität Erlangen-Nürnberg (DE) - **Topics:** Remote-Sensing Image Classification; Advanced Neural Network Applications; Remote Sensing and Land Use - **Keywords:** semantic segmentation; remote sensing images; channel attention; multi-head self-attention - **Abstract:** Semantic segmentation of remote sensing images (RSIs) is essential for applications such as environmental monitoring, urban planning, and disaster management. Convolutional Neural Networks (CNNs) and their variants struggle to capture comprehensive spectral context for learning discriminative representations. In this paper, we propose a Spectrum-Enhanced Network (SPENet) that leverages the Frequency Transformer Block (FTB) to capture rich spectral context. FTB integrates Spectrum-Enhanced Attention (SEA) with Multi-Head Frequency Self-Attention (MH-FSA), incorporating more informative contextual cues. Specifically, SEA aggregates spectral statistics through covariance matrix normalization before applying channel-wise attention. By projecting feature maps onto the frequency domain, MH-FSA provides the network with a broader context, extending beyond the low-frequency focus of standard self-attention mechanisms. Extensive experiments on the ISPRS Potsdam and LoveDA datasets show that SPENet significantly outperforms state-of-the-art methods. Besides, the proposed SEA module notably rises average F1-score/overall accuracy/mean insert over union wiht more than 2.5/2.6%/2.3%, as demonstrated by ablation study. ### Indoor Sensing with Measurements ⚑ DE - **DOI:** [10.1109/icassp49660.2025.10888550](https://doi.org/10.1109/icassp49660.2025.10888550) - **Metadaten:** Erschienen: 2025-03-12 · S. 1-5 · OpenAlex seit 2025-10-10 - **DE-Institutionen:** Ericsson (Germany) - **Autoren:** - Vijaya Yajnanarayana — Ericsson Research,India - Philipp Geuer — Ericsson (Germany) (DE) - Satyam Dwivedi — Ericsson (Sweden) (SE) - **Topics:** Indoor and Outdoor Localization Technologies; Radar Systems and Signal Processing; Millimeter-Wave Propagation and Modeling - **Keywords:** indoor sensing; cellular wireless networks; indoor channel measurements; human detection; human position estimation; convolutional neural networks - **Abstract:** The cellular wireless networks are evolving towards acquiring newer capabilities, such as sensing, which will support novel use cases and applications. Many of these require indoor sensing capabilities, which can be realized by exploiting the perturbation in the indoor channel. In this work, we conduct an indoor channel measurement campaign to study these perturbations and develop AI-based algorithms for estimating sensing parameters. We develop several AI methods based on convolutional neural networks (CNNs) and tree-based ensemble architectures for sensing. We show that the presence of a passive target like a person can be detected from the channel perturbation of a single link with more than 90 % accuracy with a simple CNN based AI algorithm. However, sensing the position of a passive target is far more challenging requiring more complex AI algorithms and deployments. We show that the position of the human in the indoor room can be estimated within the average position error of 0.7 m with a deployment having three links and employing complex AI architecture for position estimation. We also compare the results with the baseline algorithm to demonstrate the utility of the proposed method. ### Diffusion-based Identity-Preserving Facial Privacy Protection ⚑ DE - **DOI:** [10.1109/icassp49660.2025.10890829](https://doi.org/10.1109/icassp49660.2025.10890829) - **Metadaten:** Erschienen: 2025-03-12 · S. 1-5 · OpenAlex seit 2025-10-10 - **DE-Institutionen:** Friedrich Schiller University Jena; Huawei German Research Center - **Autoren:** - Dong Ki Han — Friedrich Schiller University Jena (DE) - Salaheldin Mohamed — Huawei German Research Center (DE) - Yong Li — Huawei German Research Center (DE) - Joachim Denzler — Friedrich Schiller University Jena (DE) - **Topics:** Face recognition and analysis; Generative Adversarial Networks and Image Synthesis; Digital Media Forensic Detection - **Keywords:** facial privacy protection; identity preservation; diffusion models; face recognition robustness; face verification; black-box attacks; CelebA-HQ - **Abstract:** The efficacy of facial recognition systems that utilize deep learning techniques has led to significant concerns over privacy, since they possess the capability to facilitate unauthorized monitoring of individuals in the digital realm. Current techniques for improving privacy are ineffective in producing "naturalistic" photographs that can safeguard facial features and fail to ensure privacy while maintaining an optimal user experience. We present an innovative text-agnostic method for protecting facial privacy. Our method depends on manipulating the sampling process of a pretrained diffusion model utilizing the guidance from a target image face together with the original image and face guidance in an adversarial manner to produce a protected face image. We preserve the original visual information from the input face image for identity preservation while extracting general embedding information from the target face image for soft facial attribute transfer. The output protected face image from our method has imperceptible facial changes with enhanced privacy protection against state-of-the-art (SOTA) face recognition (FR) systems. Our extensive studies have shown that the faces generated using our method have a higher level of black-box adaptability, resulting in an absolute improvement of 6.4% on CelebA-HQ compared to the current most effective SOTA facial privacy protection technique in the face verification task while maintaining high image fidelity. ### Low-Complexity Own Voice Reconstruction for Hearables with an In-Ear Microphone ⚑ DE - **DOI:** [10.1109/icassp49660.2025.10887874](https://doi.org/10.1109/icassp49660.2025.10887874) - **Metadaten:** Erschienen: 2025-03-12 · S. 1-5 · OpenAlex seit 2025-10-10 - **DE-Institutionen:** Fraunhofer Institute for Digital Media Technology - **Autoren:** - Mattes Ohlenbusch — Fraunhofer Institute for Digital Media Technology (DE) - Christian Rollwage — Fraunhofer Institute for Digital Media Technology (DE) - Simon Doclo — Fraunhofer Institute for Digital Media Technology (DE) - **Topics:** Speech and Audio Processing; Speech Recognition and Synthesis; Advanced Adaptive Filtering Techniques - **Keywords:** hearables; in-ear microphones; data augmentation; telephony; speech intelligibility; fine-tuning - **Abstract:** Hearable devices, equipped with one or more microphones, are commonly used for speech communication. Here, we consider the scenario where a hearable is used to capture the user’s own voice in a noisy environment. In this scenario, own voice reconstruction (OVR) is essential for enhancing the quality and intelligibility of the recorded noisy own voice signals for telephony applications. In previous work, we developed a deep learning-based OVR system, aiming to reduce the amount of device-specific recorded signals for training by using data augmentation with phoneme-dependent models of own voice transfer characteristics. Given the limited computational resources available on hearables, in this paper we propose low-complexity variants of an OVR system based on the frequency and time joint non-linear filter (FT-JNF) architecture and investigate the required amount of device-specific recorded signals for effective data augmentation and fine-tuning. Simulation results show that the proposed OVR system considerably improves speech quality, even under constraints of low complexity and a limited amount of device-specific recorded signals. ### Same Semantics of the Signal - What Do We Cluster with what Representation ⚑ DE - **DOI:** [10.1109/icassp49660.2025.10889674](https://doi.org/10.1109/icassp49660.2025.10889674) - **Metadaten:** Erschienen: 2025-03-12 · S. 1-5 · OpenAlex seit 2025-10-10 - **DE-Institutionen:** Friedrich-Alexander-Universität Erlangen-Nürnberg; Ostbayerische Technische Hochschule Amberg-Weiden - **Autoren:** - Alexander Barnhill — Friedrich-Alexander-Universität Erlangen-Nürnberg (DE) - Oliver Traub — Friedrich-Alexander-Universität Erlangen-Nürnberg (DE) - Andreas Maier — Friedrich-Alexander-Universität Erlangen-Nürnberg (DE) - Elmar Nöth — Friedrich-Alexander-Universität Erlangen-Nürnberg (DE) - Christian Bergler — Ostbayerische Technische Hochschule Amberg-Weiden (DE) - **Topics:** Marine animal studies overview; Music and Audio Processing; Animal Vocal Communication and Behavior - **Keywords:** semantic clustering; recording conditions; environmental noise; HuBERT; wav2vec 2.0; feature representation - **Abstract:** Semantic clustering of bioacoustic signals is crucial for a deeper understanding of intra-class differences. This is particularly important for understanding killer whale signals, as their vocalizations are learned behaviors and determination of matrilineal-specific dialects is reliant upon subtle differences within instances which may be characterized into a single larger category. Aspects of data collection may have an effect on how these calls are grouped, and it is therefore necessary to understand what the focus of the feature generation algorithm is. This study addresses the impact of factors such as recording conditions and environment, together with its respective relative noise levels, by first analyzing two different deep learning and data-driven feature representations, either derived by an undercomplete autoencoder or a supervised call type classifier. These are then compared with representations generated by two state-of-the-art transformer-based tools, namely HuBERT and Wav2Vec2. ### Speech Separation for Low-Resource Languages ⚑ DE - **DOI:** [10.1109/icassp49660.2025.10888872](https://doi.org/10.1109/icassp49660.2025.10888872) - **Metadaten:** Erschienen: 2025-03-12 · S. 1-5 · OpenAlex seit 2025-10-10 - **DE-Institutionen:** University of Bremen; FernUniversität in Hagen - **Autoren:** - Marvin Borsdorf — University of Bremen (DE) - Zexu Pan — Alibaba Group (China) (CN) - Pascal Himmelmann — FernUniversität in Hagen (DE) - Haizhou Li — Chinese University of Hong Kong (HK); Shenzhen Research Institute of Big Data (CN); Chinese University of Hong Kong, Shenzhen (CN) - Tanja Schultz — University of Bremen (DE) - **Topics:** Speech and Audio Processing; Speech Recognition and Synthesis; Music and Audio Processing - **Keywords:** speech separation; low-resource languages; blind source separation; cocktail party problem; data scarcity; multilingual training; fine-tuning; dynamic mixing - **Abstract:** Speech separation aims to equip machines with the human ability of selective listening, i.e. to focus attention on specific information in spoken communication. Studies have shown that the language spoken in a cocktail party scenario matters. While the development of speech separation models can leverage extensive databases, for the majority of languages only very limited data is available. This work presents the very first study on speech separation for low-resource languages. We choose blind source separation as the task to be studied and analyze three strategies to overcome the data scarcity of two low-resource languages from the GlobalPhoneMS2 database. We show that data from other languages can be used to develop models that work for low-resource languages. Finetuning additionally boosts the performance, and training on multiple languages increases both performance and robustness. We show that dynamic mixing in the development helps to find a trade-off between performance and development time. ### Exploring Antenna Placement Configurations with a Radar-based Silent Speech Interface ⚑ DE - **DOI:** [10.1109/icassp49660.2025.10890284](https://doi.org/10.1109/icassp49660.2025.10890284) - **Metadaten:** Erschienen: 2025-03-12 · S. 1-5 · OpenAlex seit 2025-10-10 - **DE-Institutionen:** Technische Universität Dresden - **Autoren:** - João Vítor Possamai de Menezes — Technische Universität Dresden (DE) - Martin Schütze — Technische Universität Dresden (DE) - Petr Schaffer — Technische Universität Dresden (DE) - Dirk Plettemeier — Technische Universität Dresden (DE) - Peter Birkholz — Technische Universität Dresden (DE) - **Topics:** Speech and Audio Processing; Non-Invasive Vital Sign Monitoring; Tactile and Sensory Interactions - **Keywords:** silent speech interface; antenna placement; sensor configuration; phoneme recognition; support vector machine; German speech - **Abstract:** Silent Speech Interfaces (SSIs) are systems developed to reconstruct speech based on non-acoustic bio-signals generated by brain, muscular or articulatory activity. A broad variety of sensing methods have been employed in SSIs, but determining a single optimal sensor configuration is generally a challenge. This work focuses on investigating sensing configurations for an on-body radar-based SSI and evaluating them based on their phoneme recognition performance. Audio and radar data were recorded synchronously by 8 native German speakers. One emitting and two receiving antennas were placed on the cheeks of each speaker in three different configurations, which were evaluated by nested cross-validated Support Vector Machines (SVMs). With these sensing configurations, higher accuracies than in previous works with on-body radar-based SSIs were achieved for individual speakers and for a multi-speaker evaluation. This work succeeds to produce an antenna placement guideline for future recordings with similar SSIs. ### Shabdh: A multi lingual zero-shot voice cloning approach with speaker disentanglement ⚑ DE - **DOI:** [10.1109/icassp49660.2025.10890203](https://doi.org/10.1109/icassp49660.2025.10890203) - **Metadaten:** Erschienen: 2025-03-12 · S. 1-2 · OpenAlex seit 2025-10-10 - **DE-Institutionen:** University of Bremen - **Autoren:** - Sreeram Manghat — University of Bremen (DE) - Sreeja Manghat — University of Bremen (DE) - Tanja Schultz — University of Bremen (DE) - **Topics:** Speech Recognition and Synthesis; Speech and Audio Processing; Voice and Speech Disorders - **Keywords:** speaker disentanglement; multilingual speech synthesis; neural speech codec; timbre control; pitch control; rhythm control; personalized speech synthesis; clustering - **Abstract:** This paper presents a zero-shot voice cloning system leveraging the DIS-Vector framework, which disentangles and encodes key speech features: content, pitch, timbre, and rhythm. Using the YourTTS architecture, the system synthesizes high-quality speech with precise control over both speaker identity and speech characteristics. The approach integrates multilingual data from the LIMMITS-25 dataset. The system employs neural codec TTS and clustering techniques for efficient and personalized speech synthesis. By utilizing DIS-Vector embeddings, the system enables zero-shot voice cloning, allowing the synthesis of speech in the voice of any unseen speaker with high fidelity and adaptability across multiple languages. ### RF-GML: Reference-Free Generative Machine Listener ⚑ DE - **DOI:** [10.1109/icassp49660.2025.10888107](https://doi.org/10.1109/icassp49660.2025.10888107) - **Metadaten:** Erschienen: 2025-03-12 · S. 1-5 · OpenAlex seit 2025-10-10 - **DE-Institutionen:** Rockwool Foundation Berlin - **Autoren:** - Arijit Biswas — Rockwool Foundation Berlin (DE) - Guanxin Jiang — Rockwool Foundation Berlin (DE) - **Topics:** Image and Video Quality Assessment; Speech and Audio Processing; Music and Audio Processing - **Keywords:** subjective quality prediction; coding artifacts; binaural audio; stereo audio; audio codec - **Abstract:** This paper introduces a novel reference-free (RF) audio quality metric called the RF-Generative Machine Listener (RF-GML), designed to evaluate coded mono, stereo, and binaural audio at a 48 kHz sample rate. RF-GML leverages transfer learning from a state-of-the-art full-reference (FR) Generative Machine Listener (GML) with minimal architectural modifications. The term "generative" refers to the model’s ability to generate an arbitrary number of simulated listening scores. Unlike existing RF models, RF-GML accurately predicts subjective quality scores across diverse content types and codecs. Extensive evaluations demonstrate its superiority in rating unencoded audio and distinguishing different levels of coding artifacts. RF-GML’s performance and versatility make it a valuable tool for coded audio quality assessment and monitoring in various applications, all without the need for a reference signal. ### How Machines Perceive Rooms - Regions of Relevance in Room Impulse Responses ⚑ DE - **DOI:** [10.1109/icassp49660.2025.10890456](https://doi.org/10.1109/icassp49660.2025.10890456) - **Metadaten:** Erschienen: 2025-03-12 · S. 1-5 · OpenAlex seit 2025-10-10 - **DE-Institutionen:** Technische Universität Ilmenau - **Autoren:** - Prachi Sharma — Technische Universität Ilmenau (DE) - Christian Kehling — Technische Universität Ilmenau (DE) - **Topics:** Speech and Audio Processing; Music and Audio Processing; Explainable Artificial Intelligence (XAI) - **Keywords:** room impulse response; explainable AI; layer-wise relevance propagation; room identification; acoustic parameter estimation; extended reality; direct sound; early reflections - **Abstract:** The recent surge in Extended Reality (XR) applications has sparked interest in acoustic research, particularly for enhancing immersive experiences. Room Impulse Responses (RIRs), which capture the acoustic characteristics of spaces, play a crucial role in creating plausible XR environments. Neural networks leverage RIRs for tasks such as room identification and acoustic parameter estimation. However, the decision-making process for neural networks using RIRs remains opaque. To address this, this publication applies eXplainable Artificial Intelligence (XAI) techniques, such as Layer-Wise Relevance Propagation (LRP), Linear Discriminant Analysis (LDA), and Principal Component Analysis (PCA), to identify the most relevant regions within RIRs. The techniques are applied to three independent task-dataset compositions in the area of room acoustic estimation. The findings reveal that the initial period of a RIR up to 122 ms, containing direct sound (DS) and early reflections (ER), most significantly influences network decisions. ### Classification Error Bound for Low Bayes Error Conditions in Machine Learning ⚑ DE - **DOI:** [10.1109/icassp49660.2025.10890759](https://doi.org/10.1109/icassp49660.2025.10890759) - **Metadaten:** Erschienen: 2025-03-12 · S. 1-5 · OpenAlex seit 2025-10-10 - **DE-Institutionen:** RWTH Aachen University - **Autoren:** - Zijian Győző Yang — RWTH Aachen University (DE) - Vahe Eminyan — RWTH Aachen University (DE) - Ralf Schlüter — RWTH Aachen University (DE) - Hermann Ney — RWTH Aachen University (DE) - **Topics:** Speech Recognition and Synthesis; Machine Learning and Data Classification; Machine Learning and Algorithms - **Keywords:** classification error bounds; Bayes error; Kullback-Leibler divergence; linearization; class priors; sequence classification; cross-entropy loss; perplexity - **Abstract:** In statistical classification and machine learning, classification error is an important performance measure, which is minimized by the Bayes decision rule. In practice, the unknown true distribution is usually replaced with a model distribution estimated from the training data in the Bayes decision rule. This substitution introduces a mismatch between the Bayes error and the model-based classification error. In this work, we apply classification error bounds to study the relationship between the error mismatch and the Kullback-Leibler divergence in machine learning. Motivated by recent observations of low model-based classification errors in many machine learning tasks, bounding the Bayes error to be lower, we propose a linear approximation of the classification error bound for low Bayes error conditions. Then, the bound for class priors are discussed. Moreover, we extend the classification error bound for sequences. Using automatic speech recognition as a representative example of machine learning applications, this work analytically discusses the correlations among different performance measures with extended bounds, including cross-entropy loss, language model perplexity, and word error rate. ## Advanced Functional Materials (AFM) — 1 neu · _materials_ ### Self‐Powered, Ultra‐Sensitive InGaAs/ITO Nanowire Radial Junction Infrared Photodiode for Low Flux Imaging at Room Temperature ⚑ DE - **DOI:** [10.1002/adfm.78931](https://doi.org/10.1002/adfm.78931) - **Metadaten:** Erschienen: 2026-10-11 · OpenAlex seit 2026-10-12 - **DE-Institutionen:** Friedrich Schiller University Jena - **Autoren:** - Kosala Dhanawansha — Australian Research Council (AU); ARC Centre of Excellence for Transformative Meta-Optical Systems (AU) - Chaohao Chen — Australian Research Council (AU); ARC Centre of Excellence for Transformative Meta-Optical Systems (AU) - Zhe Li — Australian Research Council (AU); ARC Centre of Excellence for Transformative Meta-Optical Systems (AU) - Tuomas Haggrén — Australian Research Council (AU); ARC Centre of Excellence for Transformative Meta-Optical Systems (AU) - Dawei Liu — Australian Research Council (AU); ARC Centre of Excellence for Transformative Meta-Optical Systems (AU); Friedrich Schiller University Jena (DE) - Yang Yu — Australian Research Council (AU); ARC Centre of Excellence for Transformative Meta-Optical Systems (AU) - Baolei Liu — Beihang University (CN) - Sanjay Krishna — The Ohio State University (US) - Hark Hoe Tan — Australian Research Council (AU); ARC Centre of Excellence for Transformative Meta-Optical Systems (AU) - C. Jagadish — Australian Research Council (AU); ARC Centre of Excellence for Transformative Meta-Optical Systems (AU) - Lan Fu — Australian Research Council (AU); ARC Centre of Excellence for Transformative Meta-Optical Systems (AU) - **Topics:** Nanowire Synthesis and Applications; Advanced Semiconductor Detectors and Materials; Photonic and Optical Devices - **Abstract:** ABSTRACT Shortwave infrared (SWIR) photodetection and imaging are of increasing importance for many emerging applications, such as optical and quantum communications, night vision, biomedical imaging and sensing. While planar InGaAs and Ge‐based SWIR photodetectors are the current state‐of‐the‐art, it is challenging to achieve single‐photon level sensitivity at non‐cryogenic temperatures due to their high dark current and thermal noise. Here, we demonstrate a highly sensitive self‐powered InGaAs nanowire photodetector with single‐photon‐level sensitivity at 1310 nm. The device utilizes a p+ InGaAs/Indium Tin Oxide (ITO) radial heterojunction nanowire structure, enabling strong optical confinement and efficient carrier extraction. It demonstrates a detection sensitivity down to 118.1 pW at room temperature and 6.5 pW at 223 K, corresponding to two photons per pulse of a 1310 nm laser excitation at 20 MHz with the noise equivalent powers of 1.59 × 10 −13 W Hz −1/2 and 2.18 × 10 −15 W Hz −1/2 respectively. The photodetector maintains a high signal‐to‐noise ratio of 58 dB at 1550 nm and cut‐off frequency of 15 MHz. Also, single‐pixel imaging under illumination down to ∼580 µW cm − 2 , showcasing the robust imaging capability under extremely low‐light conditions. Our results highlight the potential of nanowire‐based heterojunction photodetectors for the development of next‐generation, low‐cost, high‐performance SWIR photodetection and imaging technology.