1,721,456 research outputs found
High performance speaker verification system based on multilayer perceptrons and real-time enrollment
No full textSpeaker verification systems based on multilayer perceptrons (MLI's) have good prospects in reliability and flexibility as required for a successful authentication system. However, poor learning speed of error backpropagation (EBP), the representative method of learning for MLPs, has been the major problem which must be resolved to achieve real-time user enrollment. In this paper, we implement an MLP-based speaker verification system by applying methods of omitting patterns in instant learning (OIL) and discriminative cohort speakers (DCS) to approach the real-time enrollment. We evaluate the system on a Korean speech database and demonstrate the feasibility of it as a speaker verification system of high performance
Motion Artifact Reduction in PPG Signals from Face: Face Tracking & Stochastic State Space Modeling Approach
No full textThe Photoplethymography(PPG) is generally measured on a finger or an ear using contact sensors. The recent several studies using non-contact sensor such as CCD camera and web-cam to measure PPG have been introduced under the desktop or mobile computing environment. However the motion artifact issue is also emerging in non-contact camera sensing similar to contact-type one because it is sensitive to artifacts generated by subject’s head and body motion. In this paper, the two sequential approaches for a motion artifact reduction algorithm are presented; the one is a face tracking method that detects and tracks the skin region of face which is containing PPG signals, the other is the reduction method of motion artifact due to various head & face movement such as roll, yaw, pitch, translation and scale. PPG signals are modeled by stochastic state space modeling(SSM) approach and its system parameters are estimated by subspace system identification. Finally, the Kalman filter(KF) built by these parameters is applied to predict and correct distorted PPG signals. Results of the proposed KF are compared to these of the FIR band pass filter(BPF)
Elastic learning rate on error backpropagation of online update
No full textThe error-backpropagation (EBP) algorithm for learning multilayer perceptrons (MLPs) is known to have good features of robustness and economical efficiency. However, the algorithm has difficulty in selecting an optimal constant learning rate and thus results in non-optimal learning speed and inflexible operation for working data. This paper introduces an elastic learning rate that guarantees convergence of learning and its local realization by online update of MLP parameters into the original EBP algorithm in order to complement the non-optimality. The results of experiments on a speaker verification system with Korean speech database are presented and discussed to demonstrate the performance improvement of the proposed method in terms of learning speed and flexibility for working data of the original EBP algorithm
Attribute-guided Relevance Propagation for interpreting image classifier based on Deep Neural Networks
No full textDeep learning techniques have emerged as powerful tools for addressing complex and varied problems, achieving remarkable success across numerous AI domains. Despite their effectiveness, the inherent complexity of deep learning models makes them considered black boxes, reducing their interpretability and reliability. To address this challenge, we propose a novel approach called Attribute-guided Relevance Propagation (ARP). ARP enhances the interpretability of deep learning models by learning attributes from specific layers within a pre-trained image classifier and integrating these attributes into saliency maps. This integration not only improves the saliency maps but also identifies and provides example images related to key regions reflected in the maps. We validate the efficacy of ARP through both quantitative and qualitative evaluations, employing widely recognized image classifiers such as ResNet-50 and ViT trained on the benchmark datasets.
- …
