A5016625595- Digital Media Forensic Detection
- Anomaly Detection Techniques and Applications
- Topic Modeling
- Bayesian Modeling and Causal Inference
- Adversarial Robustness in Machine Learning
- Natural Language Processing Techniques
- Music and Audio Processing
- Voice and Speech Disorders
- Data Management and Algorithms
- Speech Recognition and Synthesis
- Data Quality and Management
- Speech and Audio Processing
- Semantic Web and Ontologies
Dongguan University of Technology
2022-2024
Existing deep learning models for spoofing speech detection often struggle to effectively generalize unseen attacks that were not present during the training stage. Moreover, presence of class imbalance further compounds this issue by biasing process towards seen attack samples. To address these challenges, we an innovative end-to-end model called One-Class Neural Network with Directed Statistics Pooling (OCNet-DSP). Our incorporates a feature cropping operation attenuate high-frequency...
The adversarial example is an input carefully designed by the attacker to tamper with output of neural network model. emergence audio end-to-end automatic speech recognition (ASR) system results. Due high difficulty tampering ASR results, noise produced today's generation methods still easy be noticed humans and machines. In this paper, we propose method that makes difficult perceived based on time-domain restriction. This hides perturbation in part limiting time domain. Our proposed more...
Speech recognition technology has been applied to all aspects of our daily life, but it faces many security issues. One the major threats is adversarial audio examples, which may tamper results acoustic speech system (ASR). In this paper, we propose an detection framework detect examples. The method based on transformer self-attention mechanism. Spectrogram features are extracted from and divided into patches. Position information embedded then fed encoder. Experimental show that achieves...