# Neue Publikationen in Ziel-Venues – 2026-10-19 Neue Werke: 19 · davon mit DE-Affiliation: 19 Venues: 60 · abgefragt ab publication_date 2026-09-19 · letzter Lauf: 2026-10-19T12:02:13 ## IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) — 18 neu · _conference_ ### A Differential Denoising Transformer for Polyp Segmentation in Colonoscopy ⚑ DE - **DOI:** [10.1109/icassp55912.2026.11460470](https://doi.org/10.1109/icassp55912.2026.11460470) - **Metadaten:** Erschienen: 2026-04-21 · S. 8192-8196 · OpenAlex seit 2026-04-22 - **DE-Institutionen:** Friedrich-Alexander-Universität Erlangen-Nürnberg - **Autoren:** - Yao Tong — Nanjing University of Chinese Medicine (CN) - Kongfa Hu — Nanjing University of Chinese Medicine (CN) - Tao Yang — Nanjing University of Chinese Medicine (CN) - Kankan She — Nanjing University of Chinese Medicine (CN) - Caiyan Dai — Nanjing University of Chinese Medicine (CN) - André Kaup — Friedrich-Alexander-Universität Erlangen-Nürnberg (DE) - Xin Li — Hohai University (CN) - **Topics:** Colorectal Cancer Screening and Detection; Medical Image Segmentation Techniques; Image and Signal Denoising Methods - **Keywords:** polyp segmentation; colonoscopy; image denoising; natural image statistics; medical image segmentation ### Mouthing-Enhanced Multimodal Hierarchical Contrastive Learning for Gloss-Free Sign Language Translation ⚑ DE - **DOI:** [10.1109/icassp55912.2026.11463378](https://doi.org/10.1109/icassp55912.2026.11463378) - **Metadaten:** Erschienen: 2026-04-21 · S. 8787-8791 · OpenAlex seit 2026-04-22 - **DE-Institutionen:** University of Göttingen - **Autoren:** - Wenfang Wu — University of Göttingen (DE) - Tingting Yuan — IMC University of Applied Sciences Krems (AT) - Yupeng Li — Hong Kong Baptist University (HK) - Daling Wang — Northeastern University (CN) - Xiaoming Fu — University of Göttingen (DE) - **Topics:** Hand Gesture Recognition Systems; Hearing Impairment and Communication; Multimodal Machine Learning Applications - **Keywords:** sign language translation; gloss-free translation; lip reading; multimodal contrastive learning; hierarchical contrastive learning; facial landmark detection; CLIP encoder; non-manual features - **Abstract:** Sign language translation (SLT) aims to convert sign language into spoken language. SLT interpreters rely on both manual signals and non-manual features, especially mouthings, which are essential for disambiguation. Recent work supported by large language models (LLMs) has advanced automated SLT. However, they focus almost exclusively on manual signals while neglecting mouthing cues, resulting in inaccurate or incomplete translations. In this paper, we propose SignMouth, a gloss-free SLT framework that jointly leverages hand and mouthing features. Mouthing regions are extracted via facial landmark detection and encoded using a frozen lip-reading model, while hand gesture features are obtained from a CLIP-based encoder. To enhance multimodal alignment, we introduce a hierarchical contrastive learning strategy across hand-mouthing and visual-textual levels. Experiments on PHOENIX14T and How2Sign demonstrate that SignMouth outperforms the strongest gloss-free baseline SpaMo, improving BLEU-4 from 24.32 to 24.71 on PHOENIX14T, and from 10.11 to 10.75 on How2Sign, without requiring visual encoder fine-tuning or any sign-specific pretraining. The detailed code is publicly available at https://github.com/wwf47/SignMouth. ### DNN-Based Online Source Counting Based on Spatial Generalized Magnitude Squared Coherence ⚑ DE - **DOI:** [10.1109/icassp55912.2026.11464958](https://doi.org/10.1109/icassp55912.2026.11464958) - **Metadaten:** Erschienen: 2026-04-21 · S. 22692-22696 · OpenAlex seit 2026-04-22 - **DE-Institutionen:** Carl von Ossietzky Universität Oldenburg; Hearing4all - **Autoren:** - Henri Gode — Carl von Ossietzky Universität Oldenburg (DE); Hearing4all (DE) - Simon Doclo — Carl von Ossietzky Universität Oldenburg (DE); Hearing4all (DE) - **Topics:** Speech and Audio Processing; Music and Audio Processing; Direction-of-Arrival Estimation Techniques - **Keywords:** spatial coherence; change detection; deep neural network; binaural hearing aids; reverberant environments; microphone array - **Abstract:** The number of active sound sources is a key parameter in many acoustic signal processing tasks, such as source localization, source separation, and multi-microphone speech enhancement. This paper proposes a novel method for online source counting by detecting changes in the number of active sources based on spatial coherence. The proposed method exploits the fact that a single coherent source in spatially white background noise yields high spatial coherence, whereas only noise results in low spatial coherence. By applying a spatial whitening operation, the source counting problem is reformulated as a change detection task, aiming to identify the time frames when the number of active sources changes. The method leverages the generalized magnitude-squared coherence as a measure to quantify spatial coherence, providing features for a compact neural network trained to detect source count changes framewise. Simulation results with binaural hearing aids in reverberant acoustic scenes with up to 4 speakers and background noise demonstrate the effectiveness of the proposed method for online source counting. ### Efficient Segment Anything with Depth-Aware Fusion and Limited Training Data ⚑ DE - **DOI:** [10.1109/icassp55912.2026.11464597](https://doi.org/10.1109/icassp55912.2026.11464597) - **Metadaten:** Erschienen: 2026-04-21 · S. 1731-1735 · OpenAlex seit 2026-04-22 - **DE-Institutionen:** Hochschule für Technik und Wirtschaft des Saarlandes - **Autoren:** - Yiming Zhou — Université Laval (CA) - Xuanjie Xie — Hochschule für Technik und Wirtschaft des Saarlandes (DE) - Panfeng Li — University of Michigan (US) - Albrecht Kunz — Hochschule für Technik und Wirtschaft des Saarlandes (DE) - Ahmad Osman — Hochschule für Technik und Wirtschaft des Saarlandes (DE) - Xavier Maldague — Université Laval (CA) - **Topics:** Advanced Neural Network Applications; Domain Adaptation and Few-Shot Learning; Medical Image Segmentation Techniques - **Keywords:** Segment Anything Model; RGB-D fusion; depth images; monocular depth estimation; lightweight segmentation; geometric priors - **Abstract:** Segment Anything Models (SAM) achieve impressive universal segmentation performance but require massive datasets (e.g., 11M images) and rely solely on RGB inputs. Recent efficient variants reduce computation but still depend on large-scale training. We propose a lightweight RGB-D fusion framework that augments EfficientViT-SAM with monocular depth priors. Depth maps are generated with a pretrained estimator and fused mid-level with RGB features through a dedicated depth encoder. Trained on only 11.2k samples (less than 0.1% of SA-1B), our method achieves higher accuracy than EfficientViT-SAM, showing that depth cues provide strong geometric priors for segmentation. Our results demonstrate that depth cues enable data-efficient and lightweight segmentation suitable for resource-limited scenarios. ### Interpolation-Aware Bitrate Ladder Optimization for Variable Framerate Video Streaming ⚑ DE - **DOI:** [10.1109/icassp55912.2026.11464354](https://doi.org/10.1109/icassp55912.2026.11464354) - **Metadaten:** Erschienen: 2026-04-21 · S. 8372-8376 · OpenAlex seit 2026-04-22 - **DE-Institutionen:** Brandenburg University of Technology Cottbus-Senftenberg - **Autoren:** - Amritha Premkumar — Brandenburg University of Technology Cottbus-Senftenberg (DE) - Christian Herglotz — Brandenburg University of Technology Cottbus-Senftenberg (DE) - **Topics:** Video Coding and Compression Technologies; Image and Video Quality Assessment; Advanced Data Compression Techniques - **Keywords:** frame interpolation; bit rate reduction; motion-compensated interpolation; video quality - **Abstract:** Conventional per-title encoding optimizes encoding parameters over resolution and bitrate while keeping the framerate fixed, which limits adaptability for content with diverse motion characteristics. This paper explores a dynamic framerate-aware per-title encoding framework that extends conventional encoding pipelines by incorporating framerate as an explicit optimization dimension. At a fixed resolution, we encode each video at multiple framerates and bitrates. At the decoder side, the lower-framerate encodes are temporally up-sampled to the original framerate using real-time frame interpolation methods, such as frame duplication, linear interpolation, and motion-compensated interpolation. We integrate this into a comprehensive end-to-end encoding and evaluation pipeline, comparing the resulting bitrate ladders against conventional fixed-framerate ladders. Experimental results using videos from the Inter4K dataset show that the ladder with Real-Time Intermediate Flow Estimation achieves an average of 15.37 % and 9.89 % bitrate reduction to yield the same PSNR and VMAF, respectively, while increasing the average decoding time by 14.62 %, demonstrating the effectiveness and runtime complexity-quality enhancement tradeoffs of interpolation-aware framerate adaptation for streaming. ### Real-Time Streaming MEL Vocoding with Generative Flow Matching ⚑ DE - **DOI:** [10.1109/icassp55912.2026.11462447](https://doi.org/10.1109/icassp55912.2026.11462447) - **Metadaten:** Erschienen: 2026-04-21 · S. 14392-14396 · OpenAlex seit 2025-10-10 - **DE-Institutionen:** Universität Hamburg - **Autoren:** - Simon Welker — Universität Hamburg (DE) - Tal Peer — Universität Hamburg (DE) - Timo Gerkmann — Universität Hamburg (DE) - **Topics:** Speech Recognition and Synthesis; Speech and Audio Processing; Music and Audio Processing - **Keywords:** real-time sound synthesis; text-to-speech - **Abstract:** The task of Mel vocoding, i.e., the inversion of a Mel magnitude spectrogram to an audio waveform, is still a key component in many text-to-speech (TTS) systems today. Based on generative flow matching, our prior work on generative STFT phase retrieval (DiffPhase), and the pseudoinverse operator of the Mel filterbank, we develop MelFlow, a streaming-capable generative Mel vocoder for speech sampled at 16 kHz with an algorithmic latency of only 32 ms and a total latency of 48 ms. We show real-time streaming capability at this latency not only in theory, but in practice on a consumer laptop GPU. Furthermore, we show that our model achieves substantially better PESQ and SI-SDR values compared to well-established not streaming-capable baselines for Mel vocoding including HiFi-GAN. ### CosAge: Federated Learning with Gradient Summaries for Centralized Client Selection ⚑ DE - **DOI:** [10.1109/icassp55912.2026.11462020](https://doi.org/10.1109/icassp55912.2026.11462020) - **Metadaten:** Erschienen: 2026-04-21 · S. 1156-1160 · OpenAlex seit 2026-04-22 - **DE-Institutionen:** Technical University of Munich; Deutsches Zentrum für Luft- und Raumfahrt e. V. (DLR) - **Autoren:** - Houman Asgari — Technical University of Munich (DE) - Stefano Rini — National Yang Ming Chiao Tung University (TW) - Andrea Munari — Deutsches Zentrum für Luft- und Raumfahrt e. V. (DLR) (DE) - **Topics:** Privacy-Preserving Technologies in Data; Stochastic Gradient Optimization Techniques; Age of Information Optimization - **Keywords:** federated learning; client selection; age of information; gradient sparsification; non-IID data; data heterogeneity - **Abstract:** Heterogeneous data and partial participation hinder the effectiveness of federated learning (FL). To compare client selection policies under a common yardstick, we adopt the Federated Learning with Gradient Summaries for Centralized Client Selection (FL-GSCCS) model, where each client transmits a lightweight gradient summary for selection and only chosen clients perform full local training with sparsified updates. Within this framework, we propose COSAGE, a hybrid centralized policy that combines Age of Information (AoI) with gradient dissimilarity computed from a proxy update via the cos4metric. Simulation results show that COSAGE consistently outperforms AoI-only and dissimilarity-only baselines in non-IID settings, and approaches the performance of clustering-based upper bounds without requiring client-to-client coordination or server access to client statistics. ### Graph-Based 3d Human Pose Estimation Using Wifi Signals ⚑ DE - **DOI:** [10.1109/icassp55912.2026.11463884](https://doi.org/10.1109/icassp55912.2026.11463884) - **Metadaten:** Erschienen: 2026-04-21 · S. 19992-19996 · OpenAlex seit 2026-04-22 - **DE-Institutionen:** RWTH Aachen University - **Autoren:** - Jichao Chen — EURECOM (FR) - Yangyang Qu — EURECOM (FR) - Ruibo Tang — RWTH Aachen University (DE) - Dirk T. M. Slock — EURECOM (FR) - **Topics:** Human Pose and Action Recognition; Indoor and Outdoor Localization Technologies; Hand Gesture Recognition Systems - **Keywords:** 3D human pose estimation; channel state information; graph convolutional networks; attention mechanism; occlusion robustness - **Abstract:** WiFi-based human pose estimation (HPE) has attracted increasing attention due to its resilience to occlusion and privacy-preserving compared to camera-based methods. However, existing WiFi-based HPE approaches often employ regression networks that directly map WiFi channel state information (CSI) to 3D joint coordinates, ignoring the inherent topological relationships among human joints. In this paper, we present GraphPose-Fi, a graph-based framework that explicitly models skeletal topology for WiFi-based 3D HPE. Our framework comprises a convolutional neural network (CNN) encoder shared across antennas for subcarrier–time feature extraction, a lightweight attention module that adaptively reweights features over time and across antennas, and a graph-based regression head that combines graph convolutional network (GCN) layers with self-attention to capture local topology and global dependencies. Our proposed method achieves state-of-the-art performance on the MM-Fi dataset in various settings. ### FLAME: Empowering Frozen LLMS for Knowledge Graph Completion ⚑ DE - **DOI:** [10.1109/icassp55912.2026.11464010](https://doi.org/10.1109/icassp55912.2026.11464010) - **Metadaten:** Erschienen: 2026-04-21 · S. 20367-20371 · OpenAlex seit 2024-09-11 - **DE-Institutionen:** Ludwig-Maximilians-Universität München - **Autoren:** - Bo Xue — Shanghai Jiao Tong University (CN) - Yi Xu — Shanghai Jiao Tong University (CN) - Bolei Ma — Munich Center for Machine Learning; Ludwig-Maximilians-Universität München (DE) - Yunchong Song — Shanghai Jiao Tong University (CN) - Jiaxin Ding — Shanghai Jiao Tong University (CN) - Luoyi Fu — Shanghai Jiao Tong University (CN) - Xinbing Wang — Shanghai Jiao Tong University (CN) - **Topics:** Advanced Graph Neural Networks; Topic Modeling; Semantic Web and Ontologies - **Keywords:** knowledge graph completion; frozen large language models; intermediate layer representations; memory efficiency - **Abstract:** Traditional knowledge graph completion (KGC) methods rely solely on structural information and struggle with sparsity, while Large Language Models (LLMs) address these limitations through rich world knowledge and strong context modeling. Fine-tuning LLMs is effective but costly, while non-fine-tuned LLMs are efficient but suboptimal. To address this trade-off, we propose FLAME, a framework that extracts context-aware hidden states from intermediate layers of frozen LLMs to train data-efficient KGC classifiers. We bridge LLM-KG semantic gaps via subgraph-based entity descriptions and employ sliced mutual information (SMI) to quantify task-relevant information in representations. Experiments demonstrate that FLAME achieves 47% improvement over non-fine-tuned LLM baselines and, to our knowledge, is the first to achieve fine-tuned performance with 188× memory efficiency and 26.11× speedup. ### Deep Unfolded Superiorized POCS for Robust Joint Transmission under Phase Misalignment ⚑ DE - **DOI:** [10.1109/icassp55912.2026.11462882](https://doi.org/10.1109/icassp55912.2026.11462882) - **Metadaten:** Erschienen: 2026-04-21 · S. 21616-21620 · OpenAlex seit 2026-04-22 - **DE-Institutionen:** Fraunhofer Institute for Telecommunications, Heinrich Hertz Institute - **Autoren:** - Maximilian Klemm — Fraunhofer Institute for Telecommunications, Heinrich Hertz Institute (DE) - Jochen Fink — Fraunhofer Institute for Telecommunications, Heinrich Hertz Institute (DE) - Ezer Melzer — Toga Networks a Huawei Company - Дорон Эзри — Toga Networks a Huawei Company - Sławomir Stańczak — Fraunhofer Institute for Telecommunications, Heinrich Hertz Institute (DE) - **Topics:** Advanced MIMO Systems Optimization; Sparse and Compressive Sensing Techniques; Advanced Wireless Communication Technologies - **Keywords:** joint transmission; phase misalignment; robust precoding; superiorization; projection onto convex sets; deep unfolding; spectral efficiency; multiple access points - **Abstract:** Joint transmission (JT) is a promising technology to increase the spectral efficiency in next-generation Wi-Fi systems by serving users coherently via multiple access points (APs) in the downlink. However, in practical systems, phase misalignment between APs can degrade the performance of JT schemes. The recently proposed robust joint transmission (ZI-RJT) scheme addresses this problem through closed-form precoders that are restricted to a certain subspace. Since the ZI-RJT precoders are derived assuming a high signal-to-noise ratio (SNR), we propose a new precoding strategy which takes into account the noise variance. We propose an iterative algorithm based on superiorization and projections onto convex sets (POCS) which aims at maximizing the spectral efficiency. Moreover, we propose a deep-unfolded variant (DU-S-POCS) of this algorithm, which achieves superior performance compared to ZI-RJT with just one iteration. ### UIDAPLE: Unsupervised Incremental Domain Adaptation through Adaptive Prompt Learning ⚑ DE - **DOI:** [10.1109/icassp49660.2025.10890768](https://doi.org/10.1109/icassp49660.2025.10890768) - **Metadaten:** Erschienen: 2025-03-12 · S. 1-5 · OpenAlex seit 2025-10-10 - **DE-Institutionen:** Goethe University Frankfurt - **Autoren:** - Samrat Mukherjee — Indian Institute of Technology Bombay (IN) - Tanuj Sur — Chennai Mathematical Institute (IN) - Saurish Seksaria — Indian Institute of Technology Bombay (IN) - Subhasis Chaudhuri — Indian Institute of Technology Bombay (IN) - Gemma Roig — Goethe University Frankfurt (DE) - Biplab Banerjee — Indian Institute of Technology Bombay (IN) - **Topics:** Domain Adaptation and Few-Shot Learning; Advanced Neural Network Applications; Data Stream Mining Techniques - **Keywords:** catastrophic forgetting; continual learning; prompt learning; CLIP; domain generalization; domain shift - **Abstract:** Continual learning poses significant challenges for deep neural networks, notably catastrophic forgetting, particularly when faced with shifting data distributions that compromise previously acquired knowledge. This paper tackles these issues within the Unsupervised Incremental Domain Adaptation (UIDA) framework, where the initial source domain is labeled, but subsequent domains are not. Existing methods often struggle with limited cross-domain generalization and adaptation capabilities. As a remedy, we introduce UIDAPLE, a novel approach that utilizes a unified prompt across all domains, leveraging the foundation model CLIP to obviate the need for isolated domain treatments. Specifically, UIDAPLE implements supervised prompt learning in the labeled source domain and extends this learning to unlabeled domains through confidence-based adaptation. We also present an efficient parameter alignment strategy that maintains semantic coherence across domains, effectively balancing stability and plasticity to combat catastrophic forgetting. Extensive evaluations on two benchmark datasets reveal that UIDAPLE markedly surpasses other UIDA techniques in performance. ### SELMA: A Speech-Enabled Language Model for Virtual Assistant Interactions ⚑ DE - **DOI:** [10.1109/icassp49660.2025.10890139](https://doi.org/10.1109/icassp49660.2025.10890139) - **Metadaten:** Erschienen: 2025-03-12 · S. 1-5 · OpenAlex seit 2025-10-10 - **DE-Institutionen:** Georg Simon Ohm University of Applied Sciences Nuremberg - **Autoren:** - Dominik Wagner — Georg Simon Ohm University of Applied Sciences Nuremberg (DE) - Alexander Churchill — Apple (United States) (US) - Siddharth Sigtia — Apple (United States) (US) - Erik Marchi — Apple (United States) (US) - **Topics:** Speech Recognition and Synthesis; Speech and dialogue systems; Speech and Audio Processing - **Keywords:** virtual assistants; multimodal input; audio-text fusion; large language models; speech recognition; low-rank adaptation; feature pooling; end-to-end model - **Abstract:** In this work, we present and evaluate SELMA, a Speech-Enabled Language Model for virtual Assistant interactions that integrates audio and text as inputs to a Large Language Model (LLM). SELMA is designed to handle three primary and two auxiliary tasks related to interactions with virtual assistants simultaneously within a single end-to-end model. We employ low-rank adaptation modules for parameter-efficient training of both the audio encoder and the LLM. Additionally, we implement a feature pooling strategy enabling the system to recognize global patterns and improve accuracy on tasks less reliant on individual sequence elements. Experimental results on Voice Trigger (VT) detection, Device-Directed Speech Detection (DDSD), and Automatic Speech Recognition (ASR), demonstrate that our approach both simplifies the typical input processing pipeline of virtual assistants significantly and also improves performance compared to dedicated models for each individual task. SELMA yields relative Equal-Error Rate improvements of 64% on the VT detection task, and 22% on DDSD, while also achieving word error rates close to the baseline. ### Acoustic Position Estimation of a Silent Listener ⚑ DE - **DOI:** [10.1109/icassp49660.2025.10888595](https://doi.org/10.1109/icassp49660.2025.10888595) - **Metadaten:** Erschienen: 2025-03-12 · S. 1-5 · OpenAlex seit 2025-10-10 - **DE-Institutionen:** Friedrich-Alexander-Universität Erlangen-Nürnberg; International Audio Laboratories Erlangen; Fraunhofer Institute for Integrated Circuits - **Autoren:** - Jeremy Lawrence — Friedrich-Alexander-Universität Erlangen-Nürnberg (DE); International Audio Laboratories Erlangen (DE) - Çağdaş Tuna — Fraunhofer Institute for Integrated Circuits (DE) - Andreas Walther — Fraunhofer Institute for Integrated Circuits (DE) - Nils Peters — Trinity College Dublin (IE) - **Topics:** Speech and Audio Processing; Direction-of-Arrival Estimation Techniques; Indoor and Outdoor Localization Technologies - **Keywords:** room impulse response estimation; microphone array; sine sweep; sound localization; room acoustics - **Abstract:** The localization of sounding objects has been extensively studied. In certain situations, however, an object of interest may not emit sound while another sound source is active. For instance, this occurs when a silent human listens to music and needs to be localized, e.g., to optimize the playback for the listener’s position. In this study, we present a method for localizing silent listeners: By continuously estimating room impulse responses using a loudspeaker emitting a known signal and a microphone array, we can detect subtle variations in the room impulse response over time, which are induced by the listener’s motion. Our results, evaluated in two different rooms, indicate that six room impulse responses, estimated using two successive 20 Hz-24 kHz sine sweeps of 1 s duration and each captured at three spatial points, are sufficient to localize a sitting listener within 35 cm accuracy 82 % and a standing listener 93 % of the time. ### Easy, Interpretable, Effective: openSMILE for voice deepfake detection ⚑ DE - **DOI:** [10.1109/icassp49660.2025.10890543](https://doi.org/10.1109/icassp49660.2025.10890543) - **Metadaten:** Erschienen: 2025-03-12 · S. 1-5 · OpenAlex seit 2025-10-10 - **DE-Institutionen:** Fraunhofer Institute for Applied and Integrated Security - **Autoren:** - Octavian Pascu — Universitatea Națională de Știință și Tehnologie Politehnica București (RO) - Dan Oneaţă — Universitatea Națională de Știință și Tehnologie Politehnica București (RO) - Horia Cucu — Universitatea Națională de Știință și Tehnologie Politehnica București (RO) - Nicolas Müller — Fraunhofer Institute for Applied and Integrated Security (DE) - **Topics:** Speech Recognition and Synthesis; Biometric Identification and Security; Speech and Audio Processing - **Keywords:** voice spoofing; openSMILE; acoustic features; interpretable features - **Abstract:** In this paper, we demonstrate that attacks in the latest ASVspoof5 dataset—a de facto standard in the field of voice authenticity and deepfake detection—can be identified with surprising accuracy using a small subset of very simplistic features. These are derived from the openSMILE library, and are scalar-valued, easy to compute, and human interpretable. For example, attack A10’s unvoiced segments have a mean length of 0.09 ± 0.02, while bona fide instances have a mean length of 0.18 ± 0.07. Using this feature alone, a threshold classifier achieves an Equal Error Rate (EER) of 10.3% for attack A10. Similarly, across all attacks, we achieve up to 0.8% EER, with an overall EER of 15.7 ± 6.0%.We explore the generalization capabilities of these features and find that some of them transfer effectively between attacks, primarily when the attacks originate from similar Text-to-Speech (TTS) architectures. This finding may indicate that voice anti-spoofing is, in part, a problem of identifying and remembering signatures or fingerprints of individual TTS systems. This allows to better understand anti-spoofing models and their challenges in real-world application. ### Robust Identifiability for Symbolic Recovery of Differential Equations ⚑ DE - **DOI:** [10.1109/icassp49660.2025.10887720](https://doi.org/10.1109/icassp49660.2025.10887720) - **Metadaten:** Erschienen: 2025-03-12 · S. 1-5 · OpenAlex seit 2025-10-10 - **DE-Institutionen:** Ludwig-Maximilians-Universität München - **Autoren:** - Hillary Hauger — Ludwig-Maximilians-Universität München (DE) - Philipp Scholl — Ludwig-Maximilians-Universität München (DE) - Gitta Kutyniok — Ludwig-Maximilians-Universität München (DE) - **Topics:** Model Reduction and Neural Networks; Control Systems and Identification; Numerical methods for differential equations - **Keywords:** identifiability; symbolic regression; partial differential equations; noise robustness; noise threshold - **Abstract:** Recent advancements in machine learning have transformed the discovery of physical laws, moving from manual derivation to data-driven methods that simultaneously learn both the structure and parameters of governing equations. This shift introduces new challenges regarding the validity of the discovered equations, particularly concerning their uniqueness and, hence, identifiability. While the issue of non-uniqueness has been well-studied in the context of parameter estimation, it remains underexplored for algorithms that recover both structure and parameters simultaneously. Early studies have primarily focused on idealized scenarios with perfect, noise-free data. In contrast, this paper investigates how noise influences the uniqueness and identifiability of physical laws governed by partial differential equations (PDEs). We develop a comprehensive mathematical framework to analyze the uniqueness of PDEs in the presence of noise and introduce new algorithms that account for noise, providing thresholds to assess uniqueness and identifying situations where excessive noise hinders reliable conclusions. Numerical experiments demonstrate the effectiveness of these algorithms in detecting uniqueness despite the presence of noise. ### The First Indoor Pathloss Radio Map Prediction Challenge ⚑ DE - **DOI:** [10.1109/icassp49660.2025.10889381](https://doi.org/10.1109/icassp49660.2025.10889381) - **Metadaten:** Erschienen: 2025-03-12 · S. 1-2 · OpenAlex seit 2025-10-10 - **DE-Institutionen:** Technische Universität Berlin - **Autoren:** - Stefanos Bakirtzis — University of Cambridge (GB) - Çağkan Yapar — Technische Universität Berlin (DE) - Kehai Qiu — University of Cambridge (GB) - Ian James Wassell — University of Cambridge (GB) - Jie Zhang — Ranplan Wireless - **Topics:** Millimeter-Wave Propagation and Modeling; Indoor and Outdoor Localization Technologies; Cognitive Radio Networks and Spectrum Sensing - **Keywords:** evaluation methodology - **Abstract:** To encourage further research and to facilitate fair comparisons in the development of deep learning-based radio propagation models, in the less explored case of directional radio signal emissions in indoor propagation environments, we have launched the ICASSP 2025 First Indoor Pathloss Radio Map Prediction Challenge. This overview paper describes the indoor path loss prediction problem, the datasets used, the Challenge tasks, and the evaluation methodology. Finally, the results of the Challenge and a summary of the submitted methods are presented. ### FLAMO: An Open-Source Library for Frequency-Domain Differentiable Audio Processing ⚑ DE - **DOI:** [10.1109/icassp49660.2025.10888532](https://doi.org/10.1109/icassp49660.2025.10888532) - **Metadaten:** Erschienen: 2025-03-12 · S. 1-5 · OpenAlex seit 2025-10-10 - **DE-Institutionen:** Friedrich-Alexander-Universität Erlangen-Nürnberg - **Autoren:** - Gloria Dal Santo — Aalto University (FI) - Gian Marco De Bortoli — Aalto University (FI) - Karolina Prawda — University of York (GB) - Sebastian J. Schlecht — Friedrich-Alexander-Universität Erlangen-Nürnberg (DE) - Vesa Välimäki — Aalto University (FI) - **Topics:** Digital Filter Design and Implementation; Speech and Audio Processing; Music and Audio Processing - **Keywords:** frequency-domain filter design; artificial reverberation; active acoustics - **Abstract:** We present FLAMO, a Frequency-sampling Library for Audio-Module Optimization designed to implement and optimize differentiable linear time-invariant audio systems. The library is open-source and built on the frequency-sampling filter design method, allowing for the creation of differentiable modules that can be used stand-alone or within the computation graph of neural networks, simplifying the development of differentiable audio systems. It includes predefined filtering modules and auxiliary classes for constructing, training, and logging the optimized systems, all accessible through an intuitive interface. Practical application of these modules is demonstrated through two case studies: the optimization of an artificial reverberator and an active acoustics system for improved response coloration. ### Right Label Context in End-to-End Training of Time-Synchronous ASR Models ⚑ DE - **DOI:** [10.1109/icassp49660.2025.10888716](https://doi.org/10.1109/icassp49660.2025.10888716) - **Metadaten:** Erschienen: 2025-03-12 · S. 1-5 · OpenAlex seit 2025-10-10 - **DE-Institutionen:** RWTH Aachen University - **Autoren:** - Tina Raissi — RWTH Aachen University (DE) - Ralf Schlüter — RWTH Aachen University (DE) - Hermann Ney — RWTH Aachen University (DE) - **Topics:** Speech Recognition and Synthesis; Speech and Audio Processing; Speech and dialogue systems - **Keywords:** sequence-to-sequence models; limited training data; switchboards; LibriSpeech - **Abstract:** Current time-synchronous sequence-to-sequence automatic speech recognition (ASR) models are trained by using sequence level cross-entropy that sums over all alignments. Due to the discriminative formulation, incorporating the right label context into the training criterion’s gradient causes normalization problems and is not mathematically well-defined. The classic hybrid neural network hidden Markov model (NN-HMM) with its inherent generative formulation enables conditioning on the right label context. However, due to the HMM state-tying the identity of the right label context is never modeled explicitly. In this work, we propose a factored loss with auxiliary left and right label contexts that sums over all alignments. We show that the inclusion of the right label context is particularly beneficial when training data resources are limited. Moreover, we also show that it is possible to build a factored hybrid HMM system by relying exclusively on the full-sum criterion. Experiments were conducted on Switchboard 300h and LibriSpeech 960h. ## Solid-State Electronics (SSE) — 1 neu · _materials_ ### SOI-based nanowire field-effect transistor biosensor with diamond-like carbon modification ⚑ DE - **DOI:** [10.1016/j.sse.2026.109439](https://doi.org/10.1016/j.sse.2026.109439) - **Metadaten:** Erschienen: 2026-10-01 · S. 109439 · OpenAlex seit 2026-10-19 - **DE-Institutionen:** Forschungszentrum Jülich; TU Dortmund University - **Autoren:** - H. Long — Forschungszentrum Jülich (DE); TU Dortmund University (DE) - Y. Zhang — Forschungszentrum Jülich (DE); TU Dortmund University (DE) - D. Zhulai — Forschungszentrum Jülich (DE) - N. Boichuk — Forschungszentrum Jülich (DE); TU Dortmund University (DE) - D. Pustovyi — Forschungszentrum Jülich (DE); TU Dortmund University (DE) - V. Chekubasheva — Forschungszentrum Jülich (DE); TU Dortmund University (DE) - S. Vitusevich — Forschungszentrum Jülich (DE); TU Dortmund University (DE) - **Topics:** Nanowire Synthesis and Applications; Analytical Chemistry and Sensors; Silicon Nanostructures and Photoluminescence