Institution profile

Wroclaw University of Technology

Academic institutioneurope · pl
Official website
Research library170linked papers
Opportunities0open roles
Selected work

Representative Papers

MLAAD: The Multi-Language Audio Anti-Spoofing Dataset

Jan 17, 2024IEEE International Joint Conference on Neural Network

Existing anti-spoofing audio detection systems suffer from poor generalization and cross-lingual robustness due to overreliance on English and Chinese data. Method: We introduce MLAAD—the first large-scale multilingual anti-spoofing audio dataset—comprising 160.2 hours of speech across 23 languages, synthesized using 52 TTS models (spanning 22 architectures). MLAAD bridges the critical gap in non-English/non-Chinese spoofing data. For the first time, we systematically overcome language bias in spoofing detection, demonstrating complementarity with ASVspoof 2019 and substantially improving cross-lingual detection performance. Contribution/Results: Extensive cross-dataset evaluation on ResNet, LCNN, and RawNet2 shows that models trained on MLAAD consistently outperform those trained on InTheWild and FakeOrReal across eight benchmarks; moreover, MLAAD-trained models achieve state-of-the-art results on four datasets, while ASVspoof 2019-trained models lead on the other four. This advances global applicability and fairness in deepfake audio detection.

34 citations3 influentialRead paper

Advancing machine fault diagnosis: A detailed examination of convolutional neural networks

Dec 19, 2024Measurement science and technology

To address critical challenges in fault diagnosis of complex mechanical systems—including limited generalizability of CNNs, poor adaptability to heterogeneous sensor signals (e.g., vibration, acoustic), and insufficient robustness under dynamic operating conditions—this paper systematically reviews the theoretical evolution and architectural advancements of CNNs in fault diagnosis. It identifies, for the first time, three pivotal technical pathways: data augmentation, transfer learning, and hybrid CNN-RNN/Transformer architectures. Through empirical evaluation across multi-source signals, we delineate CNNs’ performance boundaries, applicability domains, and intrinsic limitations. Furthermore, we establish a comprehensive diagnostic methodology that jointly ensures reliability (high accuracy, strong robustness) and foresight (cross-device generalization, few-shot learning, online adaptation). The work delivers a reusable technical selection guide and an engineering implementation framework, thereby providing both theoretical foundations and practical paradigms for industrial intelligent maintenance.

3 citationsRead paper

PLLuM: A Family of Polish Large Language Models

Nov 05, 2025

To address the scarcity and inadequate cultural adaptation of large language models (LLMs) for non-English languages, the PLLuM project introduces the first open-source, transparent, Polish-language LLM family. Methodologically: (1) it curates a high-quality, 100-billion-token Polish pretraining corpus and a dedicated instruction-following dataset; (2) it employs a Transformer-based architecture integrating pretraining, supervised fine-tuning, and preference alignment; and (3) it incorporates hybrid output correction and multi-layer safety filtering, grounded in a responsible AI governance framework. The primary contributions are: (i) the first publicly released series of open-weight PLLuM models; (ii) state-of-the-art performance on downstream tasks—including public administration—significantly surpassing existing baselines; and (iii) bridging the critical gap in Polish LLMs to advance a sovereign, trustworthy, and culturally grounded open AI ecosystem.

2 citationsRead paper

Application of context-dependent interpretation of biosignals recognition to control a bionic multifunctional hand prosthesis

Jan 01, 2024Biocybernetics and Biomedical Engineering

To address the limited degrees of freedom (DoFs) and poor robustness in myoelectric prosthetic control for upper-limb amputees, this paper proposes a context-aware biosignal decoding framework. Specifically, it explicitly models task context—including grasp targets and movement intent—and integrates it into surface electromyography (sEMG) pattern recognition via temporal modeling (LSTM/Transformer), multimodal context encoding, and adaptive transfer learning. The method significantly improves motion classification accuracy and cross-subject/cross-session generalizability. In real-user evaluations, it achieves a mean recognition accuracy of 96.2%, an end-to-end latency of <120 ms, and enables real-time online switching among eight dexterous hand gestures. This work establishes a novel paradigm for natural, robust, and high-DoF prosthetic control.

2 citationsRead paper

Note on edge expansion and modularity in preferential attachment graphs

Jan 09, 2026

This study investigates the theoretical bounds on edge expansion and modularity in preferential attachment random graphs where each new node introduces $h \geq 2$ edges, aiming to elucidate their connectivity and community structure properties. Employing probabilistic graph theory and combinatorial analysis, the work establishes novel probabilistic bounds on the edge expansion of small vertex subsets for small values of $h$, and leverages these results to derive a tighter global upper bound on modularity. These findings advance the theoretical understanding of structural evolution in dynamic networks and provide new analytical foundations for assessing network robustness and community detection.

1 citationsRead paper
Recent publications

Latest Papers